HYPERSPECTRAL remote sensing imagery provides finer

Size: px
Start display at page:

Download "HYPERSPECTRAL remote sensing imagery provides finer"

Transcription

1 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 10, OCTOBER Fusion of Hyperspectral Data Using Segmented PCT for Color Representation and Classification Vassilis Tsagaris, Member, IEEE, Vassilis Anastassopoulos, Member, IEEE, and George A. Lampropoulos, Member, IEEE Abstract Fusion of hyperspectral data is proposed by means of partitioning the hyperspectral bands into subgroups, prior to principal components transformation (PCT). The first principal component of each subgroup is employed for image visualization. The proposed approach is general, with the number of bands in each subgroup being application dependent. Nevertheless, the paper focuses on partitions with three subgroups suitable for RGB representation. One of them employs matched-filtering based on the spectral characteristics of various materials and is very promising for classification purposes. The information content of the hyperspectral bands as well as the quality of the obtained RGB images are quantitatively assessed using measures such as the correlation coefficient, the entropy, and the maximum energy minimum correlation index. The classification performance of the proposed partitioning approaches is tested using the K-means algorithm. Index Terms Color representation, hyperspectral data fusion, image classification, principal components transformation (PCT). I. INTRODUCTION HYPERSPECTRAL remote sensing imagery provides finer resolution in the spectral domain than multispectral data. With the number of bands in the hundreds instead of the tens, a new challenge arises for conventional data analysis techniques. The huge increase in the dimensionality of the vector space requires more sophisticated methods in order to analyze the data. In this work, a novel fusion scheme for efficient representation of a hyperspectral dataset in an informative color image is introduced, employing partitioning of the hyperspectral bands. Additionally, it is experimentally proved that the proposed fusion scheme is efficient for classification purposes. A great variety of fusion methods have been proposed in the literature, due to the complexity of the problem, the different types of data involved, and the different aims of each application [1] [9]. Fusion approaches are usually applied on data that differ spatially, spectrally, or temporally [8]. In general, pixellevel fusion methods aiming at improving spatial resolution can Manuscript received April 9, 2004; revised February 15, The work of the first author was supported in part by the European Social Fund (ESF), in part by the Operational Program for Educational and Vocational Training II (EPEAEK II), and in part by the Program HRAKLEITOS of the Ministry of Education and Religious Affairs, Greece. V. Tsagaris is with the Electronics Laboratory, Department of Physics, University of Patras, Patras 26500, Greece ( tsagaris@upatras.gr). V. Anastassopoulos is with the Electronics Laboratory, Department of Physics, University of Patras, 26500, Greece ( vassilis@physics.upatras.gr). G. A. Lampropoulos is with A.U.G. Signals Ltd., Toronto, ON, M5H 4E8 Canada ( lampro@augsignals.com). Digital Object Identifier /TGRS be classified into linear methods [1], nonlinear methods [2], optimization techniques [3], neural networks [4], [5], and image pyramids [6]. A significant amount of research is performed on linear fusion methods for multispectral and hyperspectral data [7] [9]. Since hyperspectral bands differ spectrally, fusion methods are used to obtain enhanced representation for visualization or classification. The color displays that have been proposed for representation of scientific data [10], [11] are mostly based on the mathematics of the image without considering the human vision system and the way that information is being perceived by the visual expert or end-user. Fusion methods that incorporate transformation of color spaces in order to assess the final color image have been proposed in [7] and [8]. The fusion approach proposed in this work is based on partitioning the hyperspectral dataset into subgroups of bands. The principal components transformation (PCT) [also referred as pincipal components analysis (PCA)] is applied to each subgroup. The derived principal components are suitable for information representation. This approach has significant advantages over the case of conventional PCT applied directly to the entire hyperspectral cube, as far as the computational load is concerned [9]. However, the type of partitioning proposed in this work is placing emphasis on the spectral properties of certain subgroups of bands or the spectral characteristics of specific surface cover types. First, the information in different spectral regions can be conveyed to different final principal components, providing the end-user with the ability to discriminate and classify objects and regions. Second, the proposed approach is computationally less demanding since computational load decreases when the number of subgroups increases. Furthermore, the RGB representation, obtained using three of the principal components, is perceptually more expressive [12] since the components of the final color image are correlated and the energy distribution is uniform as in natural color images. Finally, the proposed partitioning approaches, and especially the last one, which is based on the spectral signature of materials or objects, are promising for classification and detection tasks. In order to fully exploit the statistical characteristics of the data, an analysis based on second-order statistics is carried out. Furthermore, the information content and its distribution among the bands are studied by means of information theory. This analysis takes into account the user s needs and leads to the most appropriate partitioning of the hyperspectral bands. The proposed fusion approach gives emphasis on partitions with three subgroups of hyperspectral bands that result in expressive RGB representations. The approach in [9] is the simplest and gives /$ IEEE

2 2366 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 10, OCTOBER 2005 the best results in terms of reducing the computational complexity. The proposed approach leads to color images with the highest possible contrast in the RGB components and, consequently, in improved representation and visual discrimination of certain end-members that are present in the scene. Thispaperisorganizedasfollows. InSectionII, acompletedata description is provided while the second-order statistical criteria employed for quantitive measures along with information theory tools for the analysis of the hyperspectral data are discussed. In Section III, the dimensionality reduction is addressed focusing on the properties of the segmented PCT. The proposed fusion approaches and the proposed partitioning methods for the hyperspectral bands are presented in Section IV. The experimental results along with the evaluation of the performance of the proposed methods in classification tasks can be found in Section V. Finally, conclusions are drawn in Section VI. II. HYPERSPECTRAL DATA ANALYSIS The hyperspectral dataset employed for analysis in this work was acquired by the AVIRIS instrument. The analysis of the data is an extensive assessment about the way the information is distributed among the electromagnetic spectrum, the importance of each spectral band and its correlation with the others. Accordingly, the correlation coefficient, the maximum energy minimum correlation (MEMC) index [12], the total entropy and the mutual information are employed. A. Data Description The Airborne Visible Infrared Imaging Spectrometer (AVIRIS) is one of the first airborne systems [13] operated by the National Aeronautics and Space Administration (NASA)/Jet Propulsion Laboratory (JPL). In the first dataset the area covered is of Cuprite Mining Distinct, NV. It has been extensively used for remote sensing purposes since it has several exposed minerals of interest, including alunite, buddingtonite, dickite, illite, kaolinite, and quartz. The second hyperspectral dataset has been provided by NASA/JPL as free data for graduate research (namely San Diego 4). The selected part of the scene depicts an urban area with forest vegetation and sea. All the data are radiometrically calibrated and expressed as radiance values. During the data collection, possibly one or two of the spectrometers were not working properly. The result is that certain bands have zero reflectance but they are used in this dataset in order to have a uniform distribution of the wavelengths over the electromagnetic spectrum. B. Second-Order Statistics The high spectral resolution leads to a significant degree of correlation between the bands, which imposes a high level of spectral redundancy. This redundancy can be determined by means of the correlation coefficient, defined as (1) where are two hyperspectral bands and their mean values, respectively. The values of the correlation coefficient satisfy the relation. In order to reveal the spectral redundancy that is present in the hyperspectral dataset, the correlation coefficient was computed for all pairs of spectral bands and the results are shown in Fig. 1 as a grayscale image. The white pixels correspond to high degree of correlation while the black ones to uncorrelated bands. This representation reveals block structures of highly correlated bands [9], a fact that is used advantageously in this work for bands partitioning. C. MEMC Index In multispectral data processing it is often necessary to show the most important directions for the data to be projected on. These directions can be obtained by means of the MEMC index [12]. Actually, MEMC determines the bands which posses the maximum amount of energy (contrast) and have the minimum correlation with the other bands MEMC (2) where is the standard deviation of band and is the correlation coefficient between band and the rest of the bands. A large value of MEMC implies that this band has a high degree of contrast (energy) and low correlation with the other bands. As it is shown in Fig. 1(b), the MEMC index for the AVIRIS data takes its higher value for the bands with wavelength near to 2500 nm. This fact implies that the corresponding hyperspectral bands have high contrast and a low degree of correlation with the rest of the bands. On the other hand, MEMC can be employed to assess the final RGB components that are obtained via the transformation of the hyperspectral bands. D. Information Content of the Hyperspectral Bands Information theory is usually employed to calculate the data compression ratio [14] or to describe the information content of natural images [15]. Each hyperspectral band can be treated as a discrete random variable distributed according to a specific probability density, say. The entropy or total information [16] is defined as For a hyperspectral band, entropy describes the total amount of information contained in this band. In Fig. 1(d) the total entropy of the entire hyperspectral dataset is depicted as a function of the wavelength. It should be mentioned that the conclusions drawn from the study of the entropy are similar with those concerning the MEMC index. The information content in the last bands of the hyperspectral dataset is greater than that in the rest of them. The common information shared by two hyperspectral bands, say and can be studied by means of mutual information that is defined as (3) (4)

3 TSAGARIS et al.: FUSION OF HYPERSPECTRAL DATA 2367 Fig. 1. Hyperspectral data analysis using (a) correlation coefficient, (b) MEMC index, (c) mutual information, and (d) entropy. where is the joint probability density of the two variables. Mutual information is always a positive quantity that vanishes only if. Therefore, it can be interpreted as a measure of the statistical dependence between the variables and, and provides an alternative way for identifying the spectral redundancy between the bands. The mutual information between all pairs of hyperspectral bands is represented in Fig. 1(c) as a grayscale image with white pixels corresponding to the highest values and black pixels corresponding to zero. Mutual information reveals, more apparently than the correlation coefficient, the block structure of bands that have a high degree of statistical dependence. Consequently, it can be used in conjunction with the correlation coefficient to define these blocks. It becomes obvious from Fig. 1(c) that mutual information decreases when the distance of the spectral bands increases. This fact indicates that partitioning should lead to subgroups containing neighboring bands so that the dimensionality of the information in each subgroup is reduced. III. DIMENSIONALITY REDUCTION OF HYPERSPECTRAL DATA In remote sensing, one of the most widely used techniques for dimensionality reduction is the principal components transformation. PCT is effective in compressing information in multivariate datasets. It gives a new orthogonal vector space where the largest amount of energy is concentrated in a few components. Accordingly, the PCT can be considered as being a fusion process since the information of the original multidimensional data is transformed mainly on the first few orthogonal components of the new space. The concentration of the available information on a single band results in maximization of the variance for the pixels and the features in this band. Simultaneously, information does not degenerate, since PCT is an information preserving transformation and consequently the source data can be restored. The main drawback of PCT is its computational burden when it is applied on the complete hyperspectral cube. Furthermore, the use of PCT in the entire dataset may not reveal information that appears only in a few pixels of certain bands. For example an area that will have the biggest reflectance percentage in a certain wavelength will not be represented appropriately in a PCT of the entire hyperspectral cube. This is due to the global statistical nature of the transformation. PCT is optimal in the sense that the first principal component will have the highest contrast and thus it can be displayed as a grayscale image with the larger percentage of visual information.

4 2368 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 10, OCTOBER 2005 Usually the first three principal components convey more than 95% of the total energy of the data. However, these components are not suitable to form an RGB image (first component as red, second as green and third as blue) since the energy is not uniformly distributed and the result will not be optimal for the human visual system. Thus, the first principal component (red) will exhibit a high degree of contrast, the second (green) will display only a limited amount of the available brightness value, whilst the third one (blue) will demonstrate an even smaller range. In addition, the three components displayed as R, G, and B are totally uncorrelated and this is an assumption that does not hold for natural images [15]. Partitioning of the hyperspectral bands prior to PCT has already been suggested in [9]. The basic concept in that work is the reduction of the computational complexity, which can be achieved when the data are highly correlated. Consequently, partitioning was based on the edges of the image representation of the correlation matrix. Simultaneously, this partitioning procedure is assessed in [9], mainly for its classification performance. For this purpose Bhattacharya distance is employed for the principal components obtained and the classification assessment is carried out using different cover types. In this work, partitioning into subgroups of bands for the hyperspectral data prior to PCT is proposed, giving emphasis on the spectral properties of certain subgroups of bands, or the spectral characteristics of specific surface cover types. Thus, the reduction of the required computational effort is not of highest priority as in [9], but is assessed in each of the proposed partitioning approaches. After performing PCT at each subgroup, the first principal component is selected for further use in image representation or classification tasks. If the number of subgroups equals three, the corresponding principal components can be used for RGB visualization. An important issue is that the correlation of these RGB components is similar to correlation between the color components of a natural RGB image. Four different partitioning approaches are presented in the next section. The first of them is general and demonstrates the usefulness of partitioning based on simple spectral characteristics of the bands, especially for RGB representation. The other three approaches are based on specific criteria such as partitioning with equal energy or partitioning based on the spectral characteristics of specific objects, minerals, or surface cover types. As far as the computational load in a dimensionality reduction procedure is concerned, it is worthy mentioning that the implementation of PCT consists of two major tasks: an eigenanalysis to generate the transformation matrix (eigenvectors matrix) and a pixel-by-pixel linear transformation. The first task requires a significant amount of computational work since it requires the evaluation of the covariance matrix of the dataset and the corresponding eigenvectors matrix, which constitutes the transformation matrix. The evaluation of the covariance matrix requires multiplications and additions per pixel for a hyperspectral dataset of bands. In order to find the eigenvalues and the corresponding eigenvectors of the covariance matrix a diagonalization process is required. The total operation count for the diagonalization and the computation of a few eigenvectors is about as indicated in [17]. The pixel-by-pixel transformation is also a time-consuming process TABLE I DESCRIPTION OF SUBGROUPS and for a hyperspectral dataset of bands it requires multiplications and additions per pixel. The overall computational complexity as a function of the number of bands is. The idea of partitioning the hyperspectral bands prior to transformation decreases the computational complexity of all tasks [9]. IV. PROPOSED PARTITIONING APPROACHES The proposed partitioning methods described in this section are based on the spectral characteristics of the bands, the way that the energy or information is distributed along the electromagnetic spectrum as well as on spectral signature characteristics of various objects, minerals or earth surface cover types. These methods can give very good results on RGB image representation, especially for inspection of regions by humans. Additionally, they give color images with improved classification characteristics since information concerning only specific patterns or cover types is brought together (fused) in the final representation. A. Hyperspectral Data Partitioning In the general form, the segmented PCT scheme is applied on the complete dataset by partitioning it into subgroups with the bands in each group possessing high degree of correlation [9]. This results in substantial computational reduction. The number of bands is denoted in subgroup, respectively, with being the total number of bands. The PCT is applied separately on each subgroup and the primary components are selected for further use. In this subsection the hyperspectral bands are partitioned into eight subgroups, according to the perception of primary colors of the human visual system, as well as according to the positions of the atmospheric windows in the infrared region of the electromagnetic spectrum [13]. Specifically, the first three subgroups correspond to the blue, green, and red regions of the visible part of the electromagnetic spectrum. The exact range of each subgroup along with the corresponding number of bands can be found in Table I. In general, more than one of the principal components can be used as features for representation or classification. These features can be regrouped and the steps can be repeated until the required data reduction ratio is achieved. The energy compaction property of the PCT ensures that the first component of

5 TSAGARIS et al.: FUSION OF HYPERSPECTRAL DATA 2369 obtained RGB representation can give to each of the primary colors the electromagnetic information that possesses a special meaning for the humans, the earth reflectance or the atmospheric permeability. Fig. 2. Percentage of energy in the first principal component of each subgroup. TABLE II CORRELATION COEFFICIENT BETWEEN THE FIRST PRINCIPAL COMPONENTS OF EACH SUBGROUP FOR DATASET 1 B. Equal Subgroups The simplest approach in partitioning the bands is to divide the hyperspectral dataset into three groups of equal size (i.e., and ). This partitioning method will be referred to as PCT of equal subgroups (PCT-ES). PCT-ES covers equal regions of the electromagnetic spectrum of the hyperspectral dataset. Using this procedure the RGB image can be formed employingtheorthogonalband(eigenvector) fromeach subgroup that corresponds to the maximum eigenvalue of the group. This band willpossess themaximumpossible energy in the groupsince all the bands of the group are neighbors (i.e., their wavelengths are close) and thus are highly correlated. Moreover, the final RGB components will have a degree of correlation similar to that found in natural color images. In the PCT-ESmethod, the computational load is reduced by a factor of. Actually, this is the highest achievable reduction in the computational complexity. C. Maximum Energy Partitioning A different approach for partitioning the spectral bands is to select the size of each subgroup, so that the eigenvalues corresponding to the first principal component of the group become maximum. For this purpose, the largest eigenvalue corresponding to the first principal component of each subgroup is calculated. An iterative procedure is employed to evaluate the number of bands, in each subgroup so that each subgroup holds the largest percentage of energy. This energy can be expressed as a percentage if the eigenvalue of the first principal component corresponding to the larger eigenvalue of the entire subgroup is divided by the sum of all the eigenvalues obtained by conventional PCT for the entire hyperspectral dataset PE (5) where is the larger eigenvalue for the th subgroup. The first three subgroups have a small number of bands; hence the corresponding PE should be small compared to the rest. The results for all eight subgroups are depicted in Fig. 2. The first principal component of each subgroup can be used as a feature and the resulting eight features hold the 96.11% of the total energy. In the same time, the computational complexity, being, is reduced almost 96%. The principal components of each group can be used as RGB components to form a color representation of the hyperspectral dataset. In that case the final color image should have a high degree of correlation as in the natural color images. The correlation coefficient for the selected eight principal components can be found in Table II. A key advantage of the proposed method is that any triplet can be used to form an RGB representation and the final color image will have a degree of correlation similar to that found in natural color images. In this way, the The use of product instead of summation in (6) ensures that the eigenvalues will be more or less of the same order. Such partitioning implies that the largest amount of energy is conveyed by each subgroup in the first principal component and will be referred to as PCT of maximum eigenvalues (PCT-ME). In this way, the case of a dominating eigenvalue and consequently a dominating color in the RGB representation is avoided. The aim of this partitioning is to convey the highest amount of energy from the original hyperspectral data to the final color image while it distributes the information almost equally among the RGB components so that the final image possesses the properties of natural color images. D. Partitioning Based on Spectral Signature This third approach for segmenting the hyperspectral bands prior to PCT aims to reveal certain spectral characteristics of the hyperspectral dataset and will be referred to as PCT based on spectral signature (PCT-BSS). For each pixel, the spectral distribution of energy depends on what type of earth s surface the pixel contains. The spectral reflectance characteristic of each material of the earth surface is usually referred to as spectral signature of the specific material. The knowledge of the spectral signature for each material is of paramount importance. The proposed partitioning of the hyperspectral bands aims to obtain (6)

6 2370 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 10, OCTOBER 2005 Fig. 3. Spectral signatures of alunite (red line), buddingtonite (green line), and kaolinite (blue line). a final RGB image in which the materials will be distinguishable if their signatures differ. Part of the motivation for high spectral resolution sensors such as AVIRIS is that the bands are narrow enough to coincide with absorption features of individual molecules. As far as the region of exposed material on the surface is comparable to the size of a pixel, such absorption features can provide the means for labeling pixels. The area covered by the available hyperspectral dataset is of Cuprite mining where several exposed minerals of interest, including alunite, buddingtonite, and kaolinite can be found [18], [19]. In this work, the spectral libraries for the materials of interest were obtained from the U.S. Geological Survey ( In order to enhance the visual representation of a known end member a matched-filtering technique is employed. For this purpose, the spectral signatures of alunite, buddingtonite and kaolinite are plotted in Fig. 3. In the proposed partitioning procedure, each of the three materials will be mainly represented in one of the final RGB components. Specifically, the spectral signature of each material is used as a matched-filter to process only the hyperspectral bands in which this specific spectral signature has the highest value among the three. Thereby, the spectral response of the specific material is maximized while the response of the composite unknown background is suppressed [20]. Thus, three subgroups of bands are formed with maximized energy for a certain material in each subgroup. The same partitioning approach has been used for classification purposes. Accordingly, a different AVIRIS dataset from San Diego was used, since it contains calm sea, manmade regions as well as vegetation that presents the general spectral signature of chlorophyll. The single band (principal component in each partition) separability between the aforementioned three classes is measured by means of the Bhattacharya distance in the way it was applied in [9]. Simultaneously, classification is carried out for the same classes using the partitioning procedure described in [9] as well. The experimental results in the next section support the conclusion that the selection of the PCT-BSS partitioning approach is outperforming for the specific classification problem. V. EXPERIMENTS AND RESULTS The methodology of partitioning the hyperspectral bands was explained thoroughly in the previous section in its general form. An example of the proposed partitioning with eight different subgroups was employed, and comparisons with the conventional PCT were carried out in terms of energy preserving and computational complexity. In this way, the dimensionality reduction achieved by the general partitioning method was established. In the subsequent paragraphs partitioning methods, primarily aiming to color representation, are implemented for the Cuprite dataset and the final RGB representations of the hyperspectral data are examined. The measures, provided in Section II, for the analysis of hyperspectral data, are used for objective evaluation of the results along with classification accuracy scores. Moreover, the dimensionality reduction provided by the partitioning approach is compared with the one obtained using the conventional PCT. A. Partitioning Using Equal Subgroups The simplest approach in segmenting the bands of the hyperspectral dataset is to derive three subgroups with equal number of bands. Such an approach reduces significantly the computational load and provides a method for color RGB representation of the hyperspectral data. In this work, the AVIRIS dataset is partitioned in three subgroups, and the first principal component of each subgroup is used as a component of the final RGB image. Thus, the electromagnetic spectrum covered by the AVIRIS instrument is divided into three equal subregions and each of them gives one component to the final RGB image. The first subregion ranges from nm, the second from nm, and the third from nm. In Fig. 4(a) a false-color composite representation of bands 183, 193, and 207 is shown, without any histogram stretching techniques that will distort the color balance of the scene. This combination of bands is used in ENVI Tutorials in order to enhance mineralogical differences. The RGB representation formed by the first principal component of each subgroup is presented in Fig. 4(b). The proposed method results in a color image with enhanced perceivable information without distorted colors. B. Partitioning for Maximum Energy The second method for partitioning of bands prior to PCT is the PCT-ME. The three subgroups of bands, derived from the Cuprite AVIRIS dataset, are chosen in order to provide the maximum product of the largest eigenvalues. The first subgroup covers the region from nm and consists of the first 80 bands. The next 55 bands comprise the second subgroup, which covers the region nm in the electromagnetic spectrum. The third subgroup contains the largest number of bands, namely 89, and ranges from nm. The first principal component of each subgroup is used as a component of the final RGB image. In the PCT-ME method 96.2% of the total energy is conveyed in the final color image by the first principal components of the subgroups. This percentage guarantees that the final color representation of the hyperspectral dataset is meaningful and in the same time suitable for the human visual system.

7 TSAGARIS et al.: FUSION OF HYPERSPECTRAL DATA 2371 Fig. 4. (a) False-color composite of the first AVIRIS dataset, (b) color representation using PCT-ES, (c) false-color composite of the second AVIRIS dataset, and (d) color representation using PCT-ES. C. Partitioning Based on Spectral Signature A partitioning method that aims to enhance the representation of certain materials, which are present in the area covered by the hyperspectral dataset, should take into consideration their spectral signature. The implementation of PCT-BSS method for the specific dataset intends on revealing alunite, kaolinite, and buddigtonite that are present in the area. The first subgroup ranges from nm and in this region of the electromagnetic spectrum the reflectance of alunite is dominant. The spectral signature of alunite is used as the transfer function of a matched-filter applied to the corresponding bands of the hyperspectral dataset. The first principal component of this subgroup is used as the blue component for the final color image. The spectral signature of kaolinite is dominating in the range from nm, as shown in Fig. 3, and the corresponding bands constitute the second subgroup after matched filtering with the spectral signature of this material. The first principal component of this subgroup is used as the green component at the final RGB image. Finally, the spectral signature of buddigtonite is used as a matched filter, in the range from nm, where its reflectance is dominating. The red component of the final RGB image is the first principal component of this subgroup. The final color image representation of the dataset is shown in Fig. 5. A comparison with the previous representation obtained by the other partitioning methods leads to the conclusion that there is a bigger variety of the primary colors (red, green, and blue) due to the matched-filtering procedure. The results of the method can be further improved if it is applied in a certain part of the image in order to maximize the response of specific materials that may be present in the area. D. Discussion on the Proposed Methods The three methods proposed in this work for RGB representation of hyperspectral data are not in competition, and data can be applied in a complementary way when different goals are to be achieved. It is noteworthy that all three methods provide RGB images having correlation properties similar to those of natural color images [12], [15]. This is because the largest principal component used from the different subgroups to form the final

8 2372 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 10, OCTOBER 2005 Fig. 5. Color representation using (a) PCT-ME first dataset, (b) PCT-BSS (alunite-blue, kaolinite-green and buddigtonite-red) first dataset, (c) PCT-ME second dataset, and (d) PCT-BSS (urban-red, vegetation-green, and sea-blue). RGB image are highly correlated. Furthermore, the PCT-ES approach provides a quite natural representation since each RGB component corresponds to equal in length hyperspectral range. The PCT-ME method gives perceptually satisfactory RGB representation due to the equal energy delivered to each of the final bands. The PCT-BSS procedure is very promising for classification purposes, since it incorporates the special spectral characteristics of the materials to be detected. In order to evaluate the performance of the proposed methods in RGB representation, one should examine the properties of the components resulted by partitioning prior to PCT and those obtained by the conventional PCT. For this purpose, the eigenvalues obtained from each subgroup are sorted in descending order, their sum is evaluated and compared with those derived from the conventional PCT. The results are plotted in Fig. 6 for the case of PCT-ME and reveal that while the cumulative eigenvalues are slightly lower compared with those from the conventional PCT for the first five components, they become very close after that point. Specifically, when the first dozen of principal components are employed the difference between PCT-ME and the conventional PCT is only 0.1%. However, Fig. 6. Comparison between cumulative eigenvalues. (Solid line) PCT-ME. (Dashed line) Conventional PCT of the entire dataset. since we are mainly interested in RGB representation, for the first three principal components this difference is 2%, with the

9 TSAGARIS et al.: FUSION OF HYPERSPECTRAL DATA 2373 TABLE III PERFORMANCE EVALUATION BASED ON ENTROPY TABLE V CORRELATION COEFFICIENT BETWEEN THE RGB COMPONENTS OF THE COLOR IMAGES OBTAINED FROM THE PROPOSED PARTITIONING APPROACHES TABLE IV PERFORMANCE EVALUATION BASED ON MEMC PCT-ME method to provide equal in energy RGB bands and perceptually excellent color image. An objective evaluation of the proposed method for color representation could be based on the measures described in Section II. For this reason the entropy and the MEMC index are calculated for each component of the final RGB image and the results for all three proposed partitioning approaches can be found in Tables III and IV, respectively. In addition the correlation coefficient between the RGB components for the three color images formed with the proposed partitioning methods is shown in Table V. The second-order statistics and the information measure calculated for the final image reveal the properties of the proposed partitioning methods. The entropy of the RGB components is significantly increased (to values greater than 4.1) compared to the values of entropy for the source hyperspectral bands, where their mean value is In this way, the information content of each component is richer and thus a meaningful representation can be obtained. As expected, the entropy is more uniformly distributed in the case of the final RGB image obtained by the PCT-ME method. Moreover, the large values of the MEMC index for the final RGB components indicate that the contrast in each component is increased and the correlation is decreased but not vanished. Obviously, the largest values for MEMC are obtained for PCT-BSS method. The way the matched-filtering approach was applied to different spectral bands results in reduced degree of correlation and thus a high MEMC value. The degree of correlation between the RGB components is similar to the correlation found in natural color images [15]. This is an advantageous property of the proposed partitioning methods because the final color image has natural colors and hues. In this way the human visual system can perceive a greater amount of information without the existence of any strange colors often imposed by histogram stretching. The values of the correlation coefficient are smaller in the case of PCT-BSS method as already explained. E. Comparison of Classification Performance In the context of evaluating the performance of the proposed partitioning methods a classification step is employed in order to highlight the applicability of the overall approach in classification problems. The classification procedure is implemented in the ENVI software and unsupervised classification is used. The well-known K-means algorithm with five different classes and five iterations is applied on the first hyperspectral dataset. The first classification approach uses all the 224 hyperspectral bands, and provides 98.2% classification accuracy. The same classification algorithm is applied to the color images formed by PCT-ES, PCT-ME, and PCT-BSS. The overall classification accuracies are 96.9%, 97.3%, and 95.9% for PCT-ES, PCT-ME, and PCT-BSS, respectively. The results reveal that satisfactory classification performance is achieved in all cases while at the same time the computational time and the number of features are reduced. The best classification score is achieved in the case of PCT-ME and this establishes the fact that maximum energy is conveyed by the source hyperspectral bands in the color representation. In the case of PCT-BSS the proposed partitioning aims to demonstrate certain materials so the comparison of this classification map is not straightforward for the case of five classes. In the case of the second hyperspectral dataset three different classes are selected visually, with the same number of pixels in

10 2374 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 43, NO. 10, OCTOBER 2005 TABLE VI CLASSIFICATION ACCURACIES FOR THE SECOND DATASET each class. These classes are urban area, forest or vegetation and calm sea. For this case, a feature selection step similar to that described in [9] is employed. Specifically, the Bhattacharya distance (BD) is calculated and only the eigenbands with BD in each subgroup, are used for classification. In this way, the computational complexity of the classification step is significantly reduced. For example, in the case of the second hyperspectral dataset, 4, 3, and 4 eigenbands are used, for the partitioning approaches PCT-ES, PCT-ME, and PCT-BSS, respectively. The classification results for each partitioning approach and each class can be found in Table VI. The simplest partitioning approach, i.e., PCT-ES, already proposed in [9], provides satisfactory classification results for all classes. The other two partitioning methods can further improve the classification performance for the urban and forest classes. In all cases the computational complexity of the classification step is reduced compared to that of applying the K-means algorithm in the entire hyperspectral cube. It must be noted that for the second hyperspectral dataset the PCT-BSS approach significantly improves the classification performance of the urban and the forest class, thus justifying the use of the matched filtering technique. VI. CONCLUSION Partitioning of the hyperspectral data cube is proposed for efficient fusion and perceptually meaningful RGB images. The obtained color images present desirable perceptual characteristics since their statistical properties are similar to those of natural color images. Dimensionality reduction is achieved through different partitioning methods after an analysis of the source hyperspectral bands. These methods provide a practical and efficient scheme for dealing with hyperspectral data for both improved representation and classification purposes. The type of the partitioning depends on the application and can be based on the number of atmospheric windows, the distribution of spectral energy or the reflectance signature of various materials. Therefore, several final images can be obtained giving different type of information for the area under inspection. Three partitioning methods were proposed for RGB representation, with perceptually significant characteristics. Specifically, the first method namely PCT-ES results in physical representation of the scene. The PCT-ME method gives images that are perceptually more comprehensive. The third method fits better for recognition of specific materials since matched filtering with the material s spectral response is used. The generality of the proposed approach is evident from the fact that each of the above three schemes proposed for RGB representation can be applied to any number of partitions. The performance evaluation based on both statistical and information measures established the method as promising. REFERENCES [1] T. Achalakul and S. Taylor, Real-time multispectral image fusion, Concurr. Computat.: Pract. Exper., vol. 13, no. 12, pp , Sep [2] S. Mukhopadhyay and B. Chanda, Fusion of 2-D grayscale images using multiscale morphology, Pattern Recognit., vol. 34, no. 10, pp , Oct [3] A. Solberg, T. Taxt, and A. Jain, A Markov random field model for classification of multisource satellite imagery, IEEE Trans. Geosci. Remote Sens., vol. 34, no. 1, pp , Jan [4] Z. Zhang, S. Suan, and F. Zheng, Image fusion based on median filters and SOFM neural networks: A three-step scheme, Signal Process., vol. 81, no. 6, pp , Jun [5] Y. V. Shkvarko, Y. S. Shmaily, R. Jaime-Rivas, and M. Torres-Cisneros, System fusion in passive sensing using a modified Hopfield network, J. Franklin Inst., vol. 338, pp , [6] Z. Liu, K. Tsukada, K. Hanasaki, Y. K. Ho, and Y. P. Dai, Image fusion by using steerable pyramid, Pattern Recognit. Lett., vol. 22, no. 9, pp , Jul [7] J. Tyo, A. Konsolakis, D. Diersen, and R. C. Olsen, Principal-components-based display strategy for spectral imagery, IEEE Trans. Geosci. Remote Sens., vol. 41, no. 3, pp , Mar [8] C. Pohl and J. L. van Genderen, Multisensor image fusion in remote sensing: Concepts, methods, and applications, Int. J. Remote Sens., vol. 19, no. 5, pp , [9] X. Jia and J. A. Richards, Segmented principal components transformation for efficient hyperspectral remote sensing image display and classification, IEEE Trans. Geosci. Remote Sens., vol. 37, no. 1, pp , Jan [10] P. K. Robertson, Visualizing color gamuts: A user interface for the effective use of perceptual color spaces in data displays, IEEE Comput. Graph. Appl., vol. 8, pp , Sep [11] P. K. Robertson and J. F. O Callagham, The generation of color sequences for univariate and bivariate mapping, IEEE Comput. Graph. Appl., vol. 6, pp , Feb [12] V. Tsagaris and V. Anastassopoulos, Multispectral image fusion for improved RGB representation based on perceptual attributes, Int. J. Remote Sens., submitted for publication. [13] G. Vane, R. Green, T. Chrien, H. Enmark, E. Hansen, and W. Porter, The airborne visible infrared imaging spectrometer, Remote Sens. Environ., vol. 44, pp , [14] J. C. A. Van der Lubbe, Information Theory. Cambridge, U.K.: Cambridge Univ. Press, [15] A. Turiel, N. Parga, D. Ruderman, and T. Cronin, Multiscaling and information content of natural color images, J. Phys. Rev. E, vol. 62, no. 1, pp , [16] T. Cover and J. Thomas, Elements of Information Theory: Wiley, [17] W. Press, S. Teukolsky, W. Vetterling, and B. Flannery, Numerical Recipes in C. Cambridge, U.K.: Cambridge Univ. Press, [18] D. Landgrebe, Signal Theory Methods in Multispectral Remote Sensing. New York: Wiley, [19] J. A. Richards and X. Jia, Remote Sensing Digital Image Analysis. Berlin, Germany: Springer-Verlag, [20] J. C. Harsanyi and C. I. Chang, Hyperspectral image classification and dimensionality reduction: An orthogonal subspace projection approach, IEEE Trans. Geosci. Remote Sens., vol. 32, no. 4, pp , Jul Vassilis Tsagaris (S 98 M 04) received the B.Sc. degree in physics and the M.Sc. degree in electronics from the University of Patras, Patras, Greece, in 1998 and 2000, respectively. He is currently pursuing the Ph.D. degree in image processing for remote sensing at the University of Patras. His main research interests include image processing, pattern recognition, remote sensing, and data fusion.

11 TSAGARIS et al.: FUSION OF HYPERSPECTRAL DATA 2375 Vassilis Anastassopoulos (M 90) received the B.Sc. degree in physics and the Ph.D. degree in electronics from the University of Patras, Patras, Greece, in 1980 and 1986, respectively. His research interests are within the scope of digital signal processing, image processing, radar signal processing, data fusion, pattern recognition, and classification. He is currently an Associate Professor at the University of Patras. He has worked for two years in Canadian academic and private research institutions and cooperates with a number of scientific-research groups worldwide. Additionally, he has been involved in various R&D programs and scientific mobility activities. He is a reviewer for a number of scientific journals and on the Editorial Board of the journal Pattern Recognition. Dr. Anastassopoulos has been a member of the Scientific, Technical, or Organizing committees of various international conferences. He has served as Associate Editor for the IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II. George A. Lampropoulos (S 80 M 81) received the B.Sc. degree from the University of Patras, Patras, Greece, the M.Sc. and Ph.D. degrees from Queen s University, Kingston, ON, Canada, in 1979, 1982, and 1985, respectively, all in electrical engineering. His specialization is in digital signal processing, mainly in electro-optical radar. He was with the Electrical Engineering Departments of the Royal Military College, Kingston ( ), Laval University ( ), and the University of Toronto (1999). He gained considerable industrial experience with SPAR Aerospace ( ) and A.U.G. Signals Ltd. (1992-today, as President and CEO). He has supervised more than 70 industrial research projects in the area of signal processing, including image registration, segmentation, detection, classification, noise reduction and restoration, identification, recognition, and fusion. He has been a leading investigator of mineral exploration programs for government and industry using hyperspectral data. He also led projects related to web-based distributed processing with the support of the NRCan GeoInnovations Program. He has published more than 200 articles in journals, conferences, and books, as well as technical reports in the area of signal processing. He has been an invited keynote speaker in major technical events (e.g., 2002 IEE annual meeting), organized and was General Chair in major multisociety conferences (i.e., International Conference on Applications of Photonic Technology or Photonics North, 1994, 1996, 1998) and Vice Chair (2000, 2002, 2003). He is member of the board in several organizations and has been included in several Who s Who including the 1989/1990 Marquis Who s Who in the World.

Multispectral image fusion for improved RGB representation based on perceptual attributes

Multispectral image fusion for improved RGB representation based on perceptual attributes International Journal of Remote Sensing Vol. 26, No. 15, 10 August 2005, 3241 3254 Multispectral image fusion for improved RGB representation based on perceptual attributes V. TSAGARIS and V. ANASTASSOPOULOS

More information

Spatially variant dimensionality reduction for the visualization of multi/hyperspectral images

Spatially variant dimensionality reduction for the visualization of multi/hyperspectral images Author manuscript, published in "International Conference on Image Analysis and Recognition, Burnaby : Canada (2011)" DOI : 10.1007/978-3-642-21593-3_38 Spatially variant dimensionality reduction for the

More information

INTELLIGENT TARGET DETECTION IN HYPERSPECTRAL IMAGERY

INTELLIGENT TARGET DETECTION IN HYPERSPECTRAL IMAGERY INTELLIGENT TARGET DETECTION IN HYPERSPECTRAL IMAGERY Ayanna Howard, Curtis Padgett, Kenneth Brown Jet Propulsion Laboratory, California Institute of Technology 4800 Oak Grove Drive, Pasadena, CA 91 109-8099

More information

Hyperspectral and Multispectral Image Fusion Using Local Spatial-Spectral Dictionary Pair

Hyperspectral and Multispectral Image Fusion Using Local Spatial-Spectral Dictionary Pair Hyperspectral and Multispectral Image Fusion Using Local Spatial-Spectral Dictionary Pair Yifan Zhang, Tuo Zhao, and Mingyi He School of Electronics and Information International Center for Information

More information

A MAXIMUM NOISE FRACTION TRANSFORM BASED ON A SENSOR NOISE MODEL FOR HYPERSPECTRAL DATA. Naoto Yokoya 1 and Akira Iwasaki 2

A MAXIMUM NOISE FRACTION TRANSFORM BASED ON A SENSOR NOISE MODEL FOR HYPERSPECTRAL DATA. Naoto Yokoya 1 and Akira Iwasaki 2 A MAXIMUM NOISE FRACTION TRANSFORM BASED ON A SENSOR NOISE MODEL FOR HYPERSPECTRAL DATA Naoto Yokoya 1 and Akira Iwasaki 1 Graduate Student, Department of Aeronautics and Astronautics, The University of

More information

ENVI Tutorial: Basic Hyperspectral Analysis

ENVI Tutorial: Basic Hyperspectral Analysis ENVI Tutorial: Basic Hyperspectral Analysis Table of Contents OVERVIEW OF THIS TUTORIAL...2 DEFINE ROIS...3 Load AVIRIS Data...3 Create and Restore ROIs...3 Extract Mean Spectra from ROIs...4 DISCRIMINATE

More information

Remote Sensed Image Classification based on Spatial and Spectral Features using SVM

Remote Sensed Image Classification based on Spatial and Spectral Features using SVM RESEARCH ARTICLE OPEN ACCESS Remote Sensed Image Classification based on Spatial and Spectral Features using SVM Mary Jasmine. E PG Scholar Department of Computer Science and Engineering, University College

More information

ENVI Classic Tutorial: Basic Hyperspectral Analysis

ENVI Classic Tutorial: Basic Hyperspectral Analysis ENVI Classic Tutorial: Basic Hyperspectral Analysis Basic Hyperspectral Analysis 2 Files Used in this Tutorial 2 Define ROIs 3 Load AVIRIS Data 3 Create and Restore ROIs 3 Extract Mean Spectra from ROIs

More information

Principal Component Image Interpretation A Logical and Statistical Approach

Principal Component Image Interpretation A Logical and Statistical Approach Principal Component Image Interpretation A Logical and Statistical Approach Md Shahid Latif M.Tech Student, Department of Remote Sensing, Birla Institute of Technology, Mesra Ranchi, Jharkhand-835215 Abstract

More information

Data: a collection of numbers or facts that require further processing before they are meaningful

Data: a collection of numbers or facts that require further processing before they are meaningful Digital Image Classification Data vs. Information Data: a collection of numbers or facts that require further processing before they are meaningful Information: Derived knowledge from raw data. Something

More information

ENVI Classic Tutorial: Multispectral Analysis of MASTER HDF Data 2

ENVI Classic Tutorial: Multispectral Analysis of MASTER HDF Data 2 ENVI Classic Tutorial: Multispectral Analysis of MASTER HDF Data Multispectral Analysis of MASTER HDF Data 2 Files Used in This Tutorial 2 Background 2 Shortwave Infrared (SWIR) Analysis 3 Opening the

More information

Hyperspectral Remote Sensing

Hyperspectral Remote Sensing Hyperspectral Remote Sensing Multi-spectral: Several comparatively wide spectral bands Hyperspectral: Many (could be hundreds) very narrow spectral bands GEOG 4110/5100 30 AVIRIS: Airborne Visible/Infrared

More information

Implementation & comparative study of different fusion techniques (WAVELET, IHS, PCA)

Implementation & comparative study of different fusion techniques (WAVELET, IHS, PCA) International Refereed Journal of Engineering and Science (IRJES) ISSN (Online) 2319-183X, (Print) 2319-1821 Volume 1, Issue 4(December 2012), PP.37-41 Implementation & comparative study of different fusion

More information

PRINCIPAL components analysis (PCA) is a widely

PRINCIPAL components analysis (PCA) is a widely 1586 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 44, NO. 6, JUNE 2006 Independent Component Analysis-Based Dimensionality Reduction With Applications in Hyperspectral Image Analysis Jing Wang,

More information

Salient Pixels and Dimensionality Reduction for Display of Multi/Hyperspectral Images

Salient Pixels and Dimensionality Reduction for Display of Multi/Hyperspectral Images Salient Pixels and Dimensionality Reduction for Display of Multi/Hyperspectral Images Steven Le Moan 1,2, Ferdinand Deger 1,2, Alamin Mansouri 1, Yvon Voisin 1,andJonY.Hardeberg 2 1 Laboratoire d Electronique,

More information

COSC160: Detection and Classification. Jeremy Bolton, PhD Assistant Teaching Professor

COSC160: Detection and Classification. Jeremy Bolton, PhD Assistant Teaching Professor COSC160: Detection and Classification Jeremy Bolton, PhD Assistant Teaching Professor Outline I. Problem I. Strategies II. Features for training III. Using spatial information? IV. Reducing dimensionality

More information

Color Local Texture Features Based Face Recognition

Color Local Texture Features Based Face Recognition Color Local Texture Features Based Face Recognition Priyanka V. Bankar Department of Electronics and Communication Engineering SKN Sinhgad College of Engineering, Korti, Pandharpur, Maharashtra, India

More information

Spectral Classification

Spectral Classification Spectral Classification Spectral Classification Supervised versus Unsupervised Classification n Unsupervised Classes are determined by the computer. Also referred to as clustering n Supervised Classes

More information

Region Based Image Fusion Using SVM

Region Based Image Fusion Using SVM Region Based Image Fusion Using SVM Yang Liu, Jian Cheng, Hanqing Lu National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences ABSTRACT This paper presents a novel

More information

Visible and Long-Wave Infrared Image Fusion Schemes for Situational. Awareness

Visible and Long-Wave Infrared Image Fusion Schemes for Situational. Awareness Visible and Long-Wave Infrared Image Fusion Schemes for Situational Awareness Multi-Dimensional Digital Signal Processing Literature Survey Nathaniel Walker The University of Texas at Austin nathaniel.walker@baesystems.com

More information

Fast Anomaly Detection Algorithms For Hyperspectral Images

Fast Anomaly Detection Algorithms For Hyperspectral Images Vol. Issue 9, September - 05 Fast Anomaly Detection Algorithms For Hyperspectral Images J. Zhou Google, Inc. ountain View, California, USA C. Kwan Signal Processing, Inc. Rockville, aryland, USA chiman.kwan@signalpro.net

More information

Copyright 2005 Center for Imaging Science Rochester Institute of Technology Rochester, NY

Copyright 2005 Center for Imaging Science Rochester Institute of Technology Rochester, NY Development of Algorithm for Fusion of Hyperspectral and Multispectral Imagery with the Objective of Improving Spatial Resolution While Retaining Spectral Data Thesis Christopher J. Bayer Dr. Carl Salvaggio

More information

Multi Focus Image Fusion Using Joint Sparse Representation

Multi Focus Image Fusion Using Joint Sparse Representation Multi Focus Image Fusion Using Joint Sparse Representation Prabhavathi.P 1 Department of Information Technology, PG Student, K.S.R College of Engineering, Tiruchengode, Tamilnadu, India 1 ABSTRACT: The

More information

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 11, NO. 4, APRIL

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 11, NO. 4, APRIL IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 11, NO. 4, APRIL 2014 753 Quality Assessment of Panchromatic and Multispectral Image Fusion for the ZY-3 Satellite: From an Information Extraction Perspective

More information

HYPERSPECTRAL REMOTE SENSING

HYPERSPECTRAL REMOTE SENSING HYPERSPECTRAL REMOTE SENSING By Samuel Rosario Overview The Electromagnetic Spectrum Radiation Types MSI vs HIS Sensors Applications Image Analysis Software Feature Extraction Information Extraction 1

More information

DEEP LEARNING TO DIVERSIFY BELIEF NETWORKS FOR REMOTE SENSING IMAGE CLASSIFICATION

DEEP LEARNING TO DIVERSIFY BELIEF NETWORKS FOR REMOTE SENSING IMAGE CLASSIFICATION DEEP LEARNING TO DIVERSIFY BELIEF NETWORKS FOR REMOTE SENSING IMAGE CLASSIFICATION S.Dhanalakshmi #1 #PG Scholar, Department of Computer Science, Dr.Sivanthi Aditanar college of Engineering, Tiruchendur

More information

MULTIVARIATE TEXTURE DISCRIMINATION USING A PRINCIPAL GEODESIC CLASSIFIER

MULTIVARIATE TEXTURE DISCRIMINATION USING A PRINCIPAL GEODESIC CLASSIFIER MULTIVARIATE TEXTURE DISCRIMINATION USING A PRINCIPAL GEODESIC CLASSIFIER A.Shabbir 1, 2 and G.Verdoolaege 1, 3 1 Department of Applied Physics, Ghent University, B-9000 Ghent, Belgium 2 Max Planck Institute

More information

Color Space Projection, Feature Fusion and Concurrent Neural Modules for Biometric Image Recognition

Color Space Projection, Feature Fusion and Concurrent Neural Modules for Biometric Image Recognition Proceedings of the 5th WSEAS Int. Conf. on COMPUTATIONAL INTELLIGENCE, MAN-MACHINE SYSTEMS AND CYBERNETICS, Venice, Italy, November 20-22, 2006 286 Color Space Projection, Fusion and Concurrent Neural

More information

A Toolbox for Teaching Image Fusion in Matlab

A Toolbox for Teaching Image Fusion in Matlab Available online at www.sciencedirect.com ScienceDirect Procedia - Social and Behavioral Sciences 197 ( 2015 ) 525 530 7th World Conference on Educational Sciences, (WCES-2015), 05-07 February 2015, Novotel

More information

DIGITAL IMAGE ANALYSIS. Image Classification: Object-based Classification

DIGITAL IMAGE ANALYSIS. Image Classification: Object-based Classification DIGITAL IMAGE ANALYSIS Image Classification: Object-based Classification Image classification Quantitative analysis used to automate the identification of features Spectral pattern recognition Unsupervised

More information

Change Detection in Remotely Sensed Images Based on Image Fusion and Fuzzy Clustering

Change Detection in Remotely Sensed Images Based on Image Fusion and Fuzzy Clustering International Journal of Electronics Engineering Research. ISSN 0975-6450 Volume 9, Number 1 (2017) pp. 141-150 Research India Publications http://www.ripublication.com Change Detection in Remotely Sensed

More information

Introduction to digital image classification

Introduction to digital image classification Introduction to digital image classification Dr. Norman Kerle, Wan Bakx MSc a.o. INTERNATIONAL INSTITUTE FOR GEO-INFORMATION SCIENCE AND EARTH OBSERVATION Purpose of lecture Main lecture topics Review

More information

Improving the Efficiency of Fast Using Semantic Similarity Algorithm

Improving the Efficiency of Fast Using Semantic Similarity Algorithm International Journal of Scientific and Research Publications, Volume 4, Issue 1, January 2014 1 Improving the Efficiency of Fast Using Semantic Similarity Algorithm D.KARTHIKA 1, S. DIVAKAR 2 Final year

More information

Classification of Hyperspectral Breast Images for Cancer Detection. Sander Parawira December 4, 2009

Classification of Hyperspectral Breast Images for Cancer Detection. Sander Parawira December 4, 2009 1 Introduction Classification of Hyperspectral Breast Images for Cancer Detection Sander Parawira December 4, 2009 parawira@stanford.edu In 2009 approximately one out of eight women has breast cancer.

More information

Design of a Dynamic Data-Driven System for Multispectral Video Processing

Design of a Dynamic Data-Driven System for Multispectral Video Processing Design of a Dynamic Data-Driven System for Multispectral Video Processing Shuvra S. Bhattacharyya University of Maryland at College Park ssb@umd.edu With contributions from H. Li, K. Sudusinghe, Y. Liu,

More information

Hyperspectral Processing II Adapted from ENVI Tutorials #14 & 15

Hyperspectral Processing II Adapted from ENVI Tutorials #14 & 15 CEE 615: Digital Image Processing Lab 14: Hyperspectral Processing II p. 1 Hyperspectral Processing II Adapted from ENVI Tutorials #14 & 15 In this lab we consider various types of spectral processing:

More information

Hydrocarbon Index an algorithm for hyperspectral detection of hydrocarbons

Hydrocarbon Index an algorithm for hyperspectral detection of hydrocarbons INT. J. REMOTE SENSING, 20 JUNE, 2004, VOL. 25, NO. 12, 2467 2473 Hydrocarbon Index an algorithm for hyperspectral detection of hydrocarbons F. KÜHN*, K. OPPERMANN and B. HÖRIG Federal Institute for Geosciences

More information

SLIDING WINDOW FOR RELATIONS MAPPING

SLIDING WINDOW FOR RELATIONS MAPPING SLIDING WINDOW FOR RELATIONS MAPPING Dana Klimesova Institute of Information Theory and Automation, Prague, Czech Republic and Czech University of Agriculture, Prague klimes@utia.cas.c klimesova@pef.czu.cz

More information

Combining Hyperspectral and LiDAR Data for Building Extraction using Machine Learning Technique

Combining Hyperspectral and LiDAR Data for Building Extraction using Machine Learning Technique Combining Hyperspectral and LiDAR Data for Building Extraction using Machine Learning Technique DR SEYED YOUSEF SADJADI, SAEID PARSIAN Geomatics Department, School of Engineering Tafresh University Tafresh

More information

Experimentation on the use of Chromaticity Features, Local Binary Pattern and Discrete Cosine Transform in Colour Texture Analysis

Experimentation on the use of Chromaticity Features, Local Binary Pattern and Discrete Cosine Transform in Colour Texture Analysis Experimentation on the use of Chromaticity Features, Local Binary Pattern and Discrete Cosine Transform in Colour Texture Analysis N.Padmapriya, Ovidiu Ghita, and Paul.F.Whelan Vision Systems Laboratory,

More information

Remote Sensing Image Analysis via a Texture Classification Neural Network

Remote Sensing Image Analysis via a Texture Classification Neural Network Remote Sensing Image Analysis via a Texture Classification Neural Network Hayit K. Greenspan and Rodney Goodman Department of Electrical Engineering California Institute of Technology, 116-81 Pasadena,

More information

Hyperspectral Image Enhancement Based on Sensor Simulation and Vector Decomposition

Hyperspectral Image Enhancement Based on Sensor Simulation and Vector Decomposition Hyperspectral Image Enhancement Based on Sensor Simulation and Vector Decomposition Ankush Khandelwal Lab for Spatial Informatics International Institute of Information Technology Hyderabad, India ankush.khandelwal@research.iiit.ac.in

More information

DESIGN OF A NOVEL IMAGE FUSION ALGORITHM FOR IMPULSE NOISE REMOVAL IN REMOTE SENSING IMAGES BY USING THE QUALITY ASSESSMENT

DESIGN OF A NOVEL IMAGE FUSION ALGORITHM FOR IMPULSE NOISE REMOVAL IN REMOTE SENSING IMAGES BY USING THE QUALITY ASSESSMENT DESIGN OF A NOVEL IMAGE FUSION ALGORITHM FOR IMPULSE NOISE REMOVAL IN REMOTE SENSING IMAGES BY USING THE QUALITY ASSESSMENT P.PAVANI, M.V.H.BHASKARA MURTHY Department of Electronics and Communication Engineering,Aditya

More information

A Robust Band Compression Technique for Hyperspectral Image Classification

A Robust Band Compression Technique for Hyperspectral Image Classification A Robust Band Compression Technique for Hyperspectral Image Classification Qazi Sami ul Haq,Lixin Shi,Linmi Tao,Shiqiang Yang Key Laboratory of Pervasive Computing, Ministry of Education Department of

More information

Object Oriented Shadow Detection and an Enhanced Method for Shadow Removal

Object Oriented Shadow Detection and an Enhanced Method for Shadow Removal Object Oriented Shadow Detection and an Enhanced Method for Shadow Removal Divya S Kumar Department of Computer Science and Engineering Sree Buddha College of Engineering, Alappuzha, India divyasreekumar91@gmail.com

More information

FOUR REDUCED-REFERENCE METRICS FOR MEASURING HYPERSPECTRAL IMAGES AFTER SPATIAL RESOLUTION ENHANCEMENT

FOUR REDUCED-REFERENCE METRICS FOR MEASURING HYPERSPECTRAL IMAGES AFTER SPATIAL RESOLUTION ENHANCEMENT In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium 00 Years ISPRS, Vienna, Austria, July 5 7, 00, IAPRS, Vol. XXXVIII, Part 7A FOUR REDUCED-REFERENCE METRICS FOR MEASURING HYPERSPECTRAL IMAGES AFTER

More information

PRINCIPAL COMPONENT ANALYSIS IMAGE DENOISING USING LOCAL PIXEL GROUPING

PRINCIPAL COMPONENT ANALYSIS IMAGE DENOISING USING LOCAL PIXEL GROUPING PRINCIPAL COMPONENT ANALYSIS IMAGE DENOISING USING LOCAL PIXEL GROUPING Divesh Kumar 1 and Dheeraj Kalra 2 1 Department of Electronics & Communication Engineering, IET, GLA University, Mathura 2 Department

More information

PCA vs. ICA Decomposition of High Resolution SAR Images: Application to Urban Structures Recognition

PCA vs. ICA Decomposition of High Resolution SAR Images: Application to Urban Structures Recognition PCA vs. ICA Decomposition of High Resolution SAR Images: Application to Urban Structures Recognition Houda Chaabouni-Chouayakh a and Mihai Datcu a,b a German Aerospace Center (DLR), Oberpfaffenhofen D-82234

More information

Visualization of High-dimensional Remote- Sensing Data Products

Visualization of High-dimensional Remote- Sensing Data Products Visualization of High-dimensional Remote- Sensing Data Products An Innovative Graduate Student Research Proposal Principle Innovator Hongqin Zhang, Graduate Student Chester F. Carlson Center for Image

More information

Engineering And Technology (affiliated to Anna University, Chennai) Tamil. Nadu, India

Engineering And Technology (affiliated to Anna University, Chennai) Tamil. Nadu, India International Journal of Advances in Engineering & Scientific Research, Vol.2, Issue 2, Feb - 2015, pp 08-13 ISSN: 2349 3607 (Online), ISSN: 2349 4824 (Print) ABSTRACT MULTI-TEMPORAL SAR IMAGE CHANGE DETECTION

More information

An Approach for Reduction of Rain Streaks from a Single Image

An Approach for Reduction of Rain Streaks from a Single Image An Approach for Reduction of Rain Streaks from a Single Image Vijayakumar Majjagi 1, Netravati U M 2 1 4 th Semester, M. Tech, Digital Electronics, Department of Electronics and Communication G M Institute

More information

Three Dimensional Motion Vectorless Compression

Three Dimensional Motion Vectorless Compression 384 IJCSNS International Journal of Computer Science and Network Security, VOL.9 No.4, April 9 Three Dimensional Motion Vectorless Compression Rohini Nagapadma and Narasimha Kaulgud* Department of E &

More information

Copyright 2005 Society of Photo-Optical Instrumentation Engineers.

Copyright 2005 Society of Photo-Optical Instrumentation Engineers. Copyright 2005 Society of Photo-Optical Instrumentation Engineers. This paper was published in the Proceedings, SPIE Symposium on Defense & Security, 28 March 1 April, 2005, Orlando, FL, Conference 5806

More information

Fusion of pixel based and object based features for classification of urban hyperspectral remote sensing data

Fusion of pixel based and object based features for classification of urban hyperspectral remote sensing data Fusion of pixel based and object based features for classification of urban hyperspectral remote sensing data Wenzhi liao a, *, Frieke Van Coillie b, Flore Devriendt b, Sidharta Gautama a, Aleksandra Pizurica

More information

ENVI Classic Tutorial: Introduction to Hyperspectral Data 2

ENVI Classic Tutorial: Introduction to Hyperspectral Data 2 ENVI Classic Tutorial: Introduction to Hyperspectral Data Introduction to Hyperspectral Data 2 Files Used in this Tutorial 2 Background: Imaging Spectrometry 4 Introduction to Spectral Processing in ENVI

More information

Spatial Information Based Image Classification Using Support Vector Machine

Spatial Information Based Image Classification Using Support Vector Machine Spatial Information Based Image Classification Using Support Vector Machine P.Jeevitha, Dr. P. Ganesh Kumar PG Scholar, Dept of IT, Regional Centre of Anna University, Coimbatore, India. Assistant Professor,

More information

An Intelligent Clustering Algorithm for High Dimensional and Highly Overlapped Photo-Thermal Infrared Imaging Data

An Intelligent Clustering Algorithm for High Dimensional and Highly Overlapped Photo-Thermal Infrared Imaging Data An Intelligent Clustering Algorithm for High Dimensional and Highly Overlapped Photo-Thermal Infrared Imaging Data Nian Zhang and Lara Thompson Department of Electrical and Computer Engineering, University

More information

Textural Features for Hyperspectral Pixel Classification

Textural Features for Hyperspectral Pixel Classification Textural Features for Hyperspectral Pixel Classification Olga Rajadell, Pedro García-Sevilla, and Filiberto Pla Depto. Lenguajes y Sistemas Informáticos Jaume I University, Campus Riu Sec s/n 12071 Castellón,

More information

Nearest Clustering Algorithm for Satellite Image Classification in Remote Sensing Applications

Nearest Clustering Algorithm for Satellite Image Classification in Remote Sensing Applications Nearest Clustering Algorithm for Satellite Image Classification in Remote Sensing Applications Anil K Goswami 1, Swati Sharma 2, Praveen Kumar 3 1 DRDO, New Delhi, India 2 PDM College of Engineering for

More information

Performance Optimizations for an Automatic Target Generation Process in Hyperspectral Analysis

Performance Optimizations for an Automatic Target Generation Process in Hyperspectral Analysis Performance Optimizations for an Automatic Target Generation Process in Hyperspectral Analysis Fernando Sierra-Pajuelo 1, Abel Paz-Gallardo 2, Centro Extremeño de Tecnologías Avanzadas C. Sola, 1. Trujillo,

More information

Linear Discriminant Analysis in Ottoman Alphabet Character Recognition

Linear Discriminant Analysis in Ottoman Alphabet Character Recognition Linear Discriminant Analysis in Ottoman Alphabet Character Recognition ZEYNEB KURT, H. IREM TURKMEN, M. ELIF KARSLIGIL Department of Computer Engineering, Yildiz Technical University, 34349 Besiktas /

More information

MULTICHANNEL image processing is studied in this

MULTICHANNEL image processing is studied in this 186 IEEE SIGNAL PROCESSING LETTERS, VOL. 6, NO. 7, JULY 1999 Vector Median-Rational Hybrid Filters for Multichannel Image Processing Lazhar Khriji and Moncef Gabbouj, Senior Member, IEEE Abstract In this

More information

Compression of RADARSAT Data with Block Adaptive Wavelets Abstract: 1. Introduction

Compression of RADARSAT Data with Block Adaptive Wavelets Abstract: 1. Introduction Compression of RADARSAT Data with Block Adaptive Wavelets Ian Cumming and Jing Wang Department of Electrical and Computer Engineering The University of British Columbia 2356 Main Mall, Vancouver, BC, Canada

More information

Learn From The Proven Best!

Learn From The Proven Best! Applied Technology Institute (ATIcourses.com) Stay Current In Your Field Broaden Your Knowledge Increase Productivity 349 Berkshire Drive Riva, Maryland 21140 888-501-2100 410-956-8805 Website: www.aticourses.com

More information

Multi-focus Image Fusion Using Stationary Wavelet Transform (SWT) with Principal Component Analysis (PCA)

Multi-focus Image Fusion Using Stationary Wavelet Transform (SWT) with Principal Component Analysis (PCA) Multi-focus Image Fusion Using Stationary Wavelet Transform (SWT) with Principal Component Analysis (PCA) Samet Aymaz 1, Cemal Köse 1 1 Department of Computer Engineering, Karadeniz Technical University,

More information

A Comparative Study of Conventional and Neural Network Classification of Multispectral Data

A Comparative Study of Conventional and Neural Network Classification of Multispectral Data A Comparative Study of Conventional and Neural Network Classification of Multispectral Data B.Solaiman & M.C.Mouchot Ecole Nationale Supérieure des Télécommunications de Bretagne B.P. 832, 29285 BREST

More information

Introduction to Remote Sensing Wednesday, September 27, 2017

Introduction to Remote Sensing Wednesday, September 27, 2017 Lab 3 (200 points) Due October 11, 2017 Multispectral Analysis of MASTER HDF Data (ENVI Classic)* Classification Methods (ENVI Classic)* SAM and SID Classification (ENVI Classic) Decision Tree Classification

More information

Facial Expression Detection Using Implemented (PCA) Algorithm

Facial Expression Detection Using Implemented (PCA) Algorithm Facial Expression Detection Using Implemented (PCA) Algorithm Dileep Gautam (M.Tech Cse) Iftm University Moradabad Up India Abstract: Facial expression plays very important role in the communication with

More information

Hyperspectral Chemical Imaging: principles and Chemometrics.

Hyperspectral Chemical Imaging: principles and Chemometrics. Hyperspectral Chemical Imaging: principles and Chemometrics aoife.gowen@ucd.ie University College Dublin University College Dublin 1,596 PhD students 6,17 international students 8,54 graduate students

More information

Image retrieval based on bag of images

Image retrieval based on bag of images University of Wollongong Research Online Faculty of Informatics - Papers (Archive) Faculty of Engineering and Information Sciences 2009 Image retrieval based on bag of images Jun Zhang University of Wollongong

More information

Texture Image Segmentation using FCM

Texture Image Segmentation using FCM Proceedings of 2012 4th International Conference on Machine Learning and Computing IPCSIT vol. 25 (2012) (2012) IACSIT Press, Singapore Texture Image Segmentation using FCM Kanchan S. Deshmukh + M.G.M

More information

BUILDING DETECTION IN VERY HIGH RESOLUTION SATELLITE IMAGE USING IHS MODEL

BUILDING DETECTION IN VERY HIGH RESOLUTION SATELLITE IMAGE USING IHS MODEL BUILDING DETECTION IN VERY HIGH RESOLUTION SATELLITE IMAGE USING IHS MODEL Shabnam Jabari, PhD Candidate Yun Zhang, Professor, P.Eng University of New Brunswick E3B 5A3, Fredericton, Canada sh.jabari@unb.ca

More information

Semi-Supervised PCA-based Face Recognition Using Self-Training

Semi-Supervised PCA-based Face Recognition Using Self-Training Semi-Supervised PCA-based Face Recognition Using Self-Training Fabio Roli and Gian Luca Marcialis Dept. of Electrical and Electronic Engineering, University of Cagliari Piazza d Armi, 09123 Cagliari, Italy

More information

International Journal of Multidisciplinary Research and Modern Education (IJMRME) ISSN (Online): ( Volume I, Issue

International Journal of Multidisciplinary Research and Modern Education (IJMRME) ISSN (Online): (  Volume I, Issue HYPERSPECTRAL IMAGE COMPRESSION USING 3D SPIHT, SPECK AND BEZW ALGORITHMS D. Muthukumar Assistant Professor in Software Systems, Kamaraj College of Engineering and Technology, Virudhunagar, Tamilnadu Abstract:

More information

A Novel Image Super-resolution Reconstruction Algorithm based on Modified Sparse Representation

A Novel Image Super-resolution Reconstruction Algorithm based on Modified Sparse Representation , pp.162-167 http://dx.doi.org/10.14257/astl.2016.138.33 A Novel Image Super-resolution Reconstruction Algorithm based on Modified Sparse Representation Liqiang Hu, Chaofeng He Shijiazhuang Tiedao University,

More information

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 12, NO. 2, FEBRUARY

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 12, NO. 2, FEBRUARY IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 12, NO. 2, FEBRUARY 2015 349 Subspace-Based Support Vector Machines for Hyperspectral Image Classification Lianru Gao, Jun Li, Member, IEEE, Mahdi Khodadadzadeh,

More information

Remote Sensing Introduction to the course

Remote Sensing Introduction to the course Remote Sensing Introduction to the course Remote Sensing (Prof. L. Biagi) Exploitation of remotely assessed data for information retrieval Data: Digital images of the Earth, obtained by sensors recording

More information

APPLICATION OF SOFTMAX REGRESSION AND ITS VALIDATION FOR SPECTRAL-BASED LAND COVER MAPPING

APPLICATION OF SOFTMAX REGRESSION AND ITS VALIDATION FOR SPECTRAL-BASED LAND COVER MAPPING APPLICATION OF SOFTMAX REGRESSION AND ITS VALIDATION FOR SPECTRAL-BASED LAND COVER MAPPING J. Wolfe a, X. Jin a, T. Bahr b, N. Holzer b, * a Harris Corporation, Broomfield, Colorado, U.S.A. (jwolfe05,

More information

Robust Phase-Based Features Extracted From Image By A Binarization Technique

Robust Phase-Based Features Extracted From Image By A Binarization Technique IOSR Journal of Computer Engineering (IOSR-JCE) e-issn: 2278-0661,p-ISSN: 2278-8727, Volume 18, Issue 4, Ver. IV (Jul.-Aug. 2016), PP 10-14 www.iosrjournals.org Robust Phase-Based Features Extracted From

More information

CS 195-5: Machine Learning Problem Set 5

CS 195-5: Machine Learning Problem Set 5 CS 195-5: Machine Learning Problem Set 5 Douglas Lanman dlanman@brown.edu 26 November 26 1 Clustering and Vector Quantization Problem 1 Part 1: In this problem we will apply Vector Quantization (VQ) to

More information

CHAPTER 4 SEMANTIC REGION-BASED IMAGE RETRIEVAL (SRBIR)

CHAPTER 4 SEMANTIC REGION-BASED IMAGE RETRIEVAL (SRBIR) 63 CHAPTER 4 SEMANTIC REGION-BASED IMAGE RETRIEVAL (SRBIR) 4.1 INTRODUCTION The Semantic Region Based Image Retrieval (SRBIR) system automatically segments the dominant foreground region and retrieves

More information

Last week. Multi-Frame Structure from Motion: Multi-View Stereo. Unknown camera viewpoints

Last week. Multi-Frame Structure from Motion: Multi-View Stereo. Unknown camera viewpoints Last week Multi-Frame Structure from Motion: Multi-View Stereo Unknown camera viewpoints Last week PCA Today Recognition Today Recognition Recognition problems What is it? Object detection Who is it? Recognizing

More information

Satellite Image Processing Using Singular Value Decomposition and Discrete Wavelet Transform

Satellite Image Processing Using Singular Value Decomposition and Discrete Wavelet Transform Satellite Image Processing Using Singular Value Decomposition and Discrete Wavelet Transform Kodhinayaki E 1, vinothkumar S 2, Karthikeyan T 3 Department of ECE 1, 2, 3, p.g scholar 1, project coordinator

More information

Multi-Sensor Fusion of Electro-Optic and Infrared Signals for High Resolution Visible Images: Part II

Multi-Sensor Fusion of Electro-Optic and Infrared Signals for High Resolution Visible Images: Part II Multi-Sensor Fusion of Electro-Optic and Infrared Signals for High Resolution Visible Images: Part II Xiaopeng Huang, Ravi Netravali*, Hong Man and Victor Lawrence Dept. of Electrical and Computer Engineering

More information

Appendix III: Ten (10) Specialty Areas - Remote Sensing/Imagry Science Curriculum Mapping to Knowledge Units-RS/Imagry Science Specialty Area

Appendix III: Ten (10) Specialty Areas - Remote Sensing/Imagry Science Curriculum Mapping to Knowledge Units-RS/Imagry Science Specialty Area III. Remote Sensing/Imagery Science Specialty Area 1. Knowledge Unit title: Remote Sensing Collection Platforms A. Knowledge Unit description and objective: Understand and be familiar with remote sensing

More information

A Framework of Hyperspectral Image Compression using Neural Networks

A Framework of Hyperspectral Image Compression using Neural Networks A Framework of Hyperspectral Image Compression using Neural Networks Yahya M. Masalmah, Ph.D 1, Christian Martínez-Nieves 1, Rafael Rivera-Soto 1, Carlos Velez 1, and Jenipher Gonzalez 1 1 Universidad

More information

Analisi di immagini iperspettrali satellitari multitemporali: metodi ed applicazioni

Analisi di immagini iperspettrali satellitari multitemporali: metodi ed applicazioni Analisi di immagini iperspettrali satellitari multitemporali: metodi ed applicazioni E-mail: bovolo@fbk.eu Web page: http://rsde.fbk.eu Outline 1 Multitemporal image analysis 2 Multitemporal images pre-processing

More information

Classify Multi-Spectral Data Classify Geologic Terrains on Venus Apply Multi-Variate Statistics

Classify Multi-Spectral Data Classify Geologic Terrains on Venus Apply Multi-Variate Statistics Classify Multi-Spectral Data Classify Geologic Terrains on Venus Apply Multi-Variate Statistics Operations What Do I Need? Classify Merge Combine Cross Scan Score Warp Respace Cover Subscene Rotate Translators

More information

A comparison study of dimension estimation algorithms

A comparison study of dimension estimation algorithms A comparison study of dimension estimation algorithms Ariel Schlamm, a Ronald G. Resmini, b David Messinger, a, and William Basener c a Rochester Institute of Technology, Center for Imaging Science, Digital

More information

A Survey on Feature Extraction Techniques for Palmprint Identification

A Survey on Feature Extraction Techniques for Palmprint Identification International Journal Of Computational Engineering Research (ijceronline.com) Vol. 03 Issue. 12 A Survey on Feature Extraction Techniques for Palmprint Identification Sincy John 1, Kumudha Raimond 2 1

More information

AN IMPROVED HYBRIDIZED K- MEANS CLUSTERING ALGORITHM (IHKMCA) FOR HIGHDIMENSIONAL DATASET & IT S PERFORMANCE ANALYSIS

AN IMPROVED HYBRIDIZED K- MEANS CLUSTERING ALGORITHM (IHKMCA) FOR HIGHDIMENSIONAL DATASET & IT S PERFORMANCE ANALYSIS AN IMPROVED HYBRIDIZED K- MEANS CLUSTERING ALGORITHM (IHKMCA) FOR HIGHDIMENSIONAL DATASET & IT S PERFORMANCE ANALYSIS H.S Behera Department of Computer Science and Engineering, Veer Surendra Sai University

More information

ADVANCED IMAGE PROCESSING METHODS FOR ULTRASONIC NDE RESEARCH C. H. Chen, University of Massachusetts Dartmouth, N.

ADVANCED IMAGE PROCESSING METHODS FOR ULTRASONIC NDE RESEARCH C. H. Chen, University of Massachusetts Dartmouth, N. ADVANCED IMAGE PROCESSING METHODS FOR ULTRASONIC NDE RESEARCH C. H. Chen, University of Massachusetts Dartmouth, N. Dartmouth, MA USA Abstract: The significant progress in ultrasonic NDE systems has now

More information

Hyperspectral Image Anomaly Targets Detection with Online Deep Learning

Hyperspectral Image Anomaly Targets Detection with Online Deep Learning This full text paper was peer-reviewed at the direction of IEEE Instrumentation and Measurement Society prior to the acceptance and publication. Hyperspectral Image Anomaly Targets Detection with Online

More information

NSCT domain image fusion, denoising & K-means clustering for SAR image change detection

NSCT domain image fusion, denoising & K-means clustering for SAR image change detection NSCT domain image fusion, denoising & K-means clustering for SAR image change detection Yamuna J. 1, Arathy C. Haran 2 1,2, Department of Electronics and Communications Engineering, 1 P. G. student, 2

More information

CHAPTER 3 PRINCIPAL COMPONENT ANALYSIS AND FISHER LINEAR DISCRIMINANT ANALYSIS

CHAPTER 3 PRINCIPAL COMPONENT ANALYSIS AND FISHER LINEAR DISCRIMINANT ANALYSIS 38 CHAPTER 3 PRINCIPAL COMPONENT ANALYSIS AND FISHER LINEAR DISCRIMINANT ANALYSIS 3.1 PRINCIPAL COMPONENT ANALYSIS (PCA) 3.1.1 Introduction In the previous chapter, a brief literature review on conventional

More information

Tensor Based Approaches for LVA Field Inference

Tensor Based Approaches for LVA Field Inference Tensor Based Approaches for LVA Field Inference Maksuda Lillah and Jeff Boisvert The importance of locally varying anisotropy (LVA) in model construction can be significant; however, it is often ignored

More information

Random projection for non-gaussian mixture models

Random projection for non-gaussian mixture models Random projection for non-gaussian mixture models Győző Gidófalvi Department of Computer Science and Engineering University of California, San Diego La Jolla, CA 92037 gyozo@cs.ucsd.edu Abstract Recently,

More information

Model-based segmentation and recognition from range data

Model-based segmentation and recognition from range data Model-based segmentation and recognition from range data Jan Boehm Institute for Photogrammetry Universität Stuttgart Germany Keywords: range image, segmentation, object recognition, CAD ABSTRACT This

More information

COMPARATIVE STUDY OF IMAGE FUSION TECHNIQUES IN SPATIAL AND TRANSFORM DOMAIN

COMPARATIVE STUDY OF IMAGE FUSION TECHNIQUES IN SPATIAL AND TRANSFORM DOMAIN COMPARATIVE STUDY OF IMAGE FUSION TECHNIQUES IN SPATIAL AND TRANSFORM DOMAIN Bhuvaneswari Balachander and D. Dhanasekaran Department of Electronics and Communication Engineering, Saveetha School of Engineering,

More information

ABSTRACT 1. INTRODUCTION

ABSTRACT 1. INTRODUCTION Correlation between lidar-derived intensity and passive optical imagery Jeremy P. Metcalf, Angela M. Kim, Fred A. Kruse, and Richard C. Olsen Physics Department and Remote Sensing Center, Naval Postgraduate

More information