Indoor Outdoor Image Classification

Size: px
Start display at page:

Download "Indoor Outdoor Image Classification"

Transcription

1 Indoor Outdoor Image Classification Bachelor Thesis Fei Guo Department of Computer Science, ETH Zurich Advisor: Lukas Bossard Supervisor: Prof. Dr. Luc van Gool August 18, 2011

2

3 Abstract Because of the increasing need for automatic photo organization, the indoor outdoor image classification as a very basic aspect of it is going to be investigated in this work. We are going to use color, texture, shape and metadata features and their combinations which could be directly extracted and inferred from images to help identify if an image shows an indoor or outdoor scene. Furthermore, we are going to divide the images into subblocks in different ways and evaluate the classification on each of them. In addition, the images will be rescaled to different resolutions and at each resolution the classification is performed. For classification, the K-Nearest Neighbor and Support Vector Machine are used. Through a series of experiments we found that the low-level feature combination of color, texture, spatial envelope gave the accuracy 78.17% and camera metadata alone achieved 79.33%.

4

5 Acknowledgements I would like to take this opportunity to sincerely thank my advisor Lukas Bossard for his patience, his help of building the image data set, labeling the ground truth and also for his weekly discussion with me. And of course, I must thank my parents who gave me financial support for the whole bachelor study in Switzerland.

6

7 Contents 1 Introduction 1 2 Background Features Color Histogram Texture Features Spatial Envelope Metadata Classification K-nearest Neighbor (KNN) Support Vector Machine (SVM) Related Work Experiments and Results Image Data Set Experiment Protocol Experiments with Single Image Features Experiments with Feature Combinations Experiments Integrated with Metadata Experiments with Different Image Dimensions Experiments with Different Division Strategies Experiments with Two Stage Support Vector Machine Conclusion 19 I

8 CONTENTS II

9 List of Figures 2.1 Two-level Wavelet Pyramid Structure Two-level Wavelet Decomposition Distribution of Flash Distribution of Exposure Time [sec] Distribution of Aperture Value Distribution of ISO Speed Rating KNN Classification SVM Separating Hyperplane Ambiguous Image Examples Histogram of Image Resolutions Color Feature and Texture Feature x3 vs. 4x Different Feature Extraction Strategies Illustration of the two-stage classification approach III

10 LIST OF FIGURES IV

11 List of Tables 3.1 Results of Single Image Features Percentage of images which have Metadata Results of Feature Combinations Results of Single Features with Metadata Results of Feature Combinations with Metadata Comparison between Normal SVM and Two Stage SVM V

12 LIST OF TABLES VI

13 Chapter 1 Introduction Nowadays, there are a lot photos being taken by different photographers at different places all over the world. Images are accumulated more and more, which raises the need for automatic photo organization. We put this work in the context of attribute based image classification. The exact content of an image must not be understood by us, e.g. a person eating in a kitchen or a car on a street. Instead, we focus on indoor/outdoor classification which can be used for organizing photo collections. Although a lot of work for image classification have been done before, we are going to select some interesting features from different works and use combinations of them. In this work, most of the relevant features are taken under consideration. Color feature can be extracted by means of histogram. Texture feature are collected by discrete wavelet decomposion. Structure of scenes will be estimated by spatial envelope. EXIF information consists of different tags, e.g. flash, exposure time, aperture value and ISO speed rating, which can be easily fetched from images. We will exploit single features, feature combinations, feature extraction strategies and different image resolutions to train different classifiers and see how these different factors affect the success rate of the classification. 1

14 CHAPTER 1. INTRODUCTION 2

15 Chapter 2 Background In general, it is very common for indoor/outdoor classification to divide images into subblocks and extract different image features from each subblock instead of the whole image. Afterwards the feature vectors for each subblock are going to be concatenated to one feature vector. In the end, the feature vectors of all images of the training set are used for training a classifier. The classifier s performance is then evaluated on the test set. 2.1 Features Many different features can be extracted from an image. In this work, we selected some interesting and important features which help us to distinguish between indoor and outdoor images Color Histogram Color histograms describe how the colors of an image are distributed. We compute the histogram of each color channel as color feature on a region of the image. According to the Benchmark for image classification [7], the Ohta color space is used for computing the color features in our case. It is a linear transformation of the RGB space. Its color channels are used by Szummer and Picard in [10] as: I1 = R + G + B I2 = R B I3 = R 2G + B In this work, we used 16-bin histograms for each channel as a feature, with in total 48 dimensions Texture Features With the texture features we are able to know the information about how pixel intensities spatially vary. The texture features are computed by means of two level wavelet decomposition as described by Serrano et al. [9]. The decomposition is performed on the I1 channel of the Ohta color space the with Daubechies 4-tap filters, where low-pass filter is: h(n) = [ ] 3

16 CHAPTER 2. BACKGROUND and the high-pass filter is: g(n) = [ ] According to the two-level wavelet pyramid structure described in their work, after the first level decomposition, all the coefficients are used for computing texture features except for the upper left one. The second decomposition is performed on the upper left part. Then, the texture features are obtained by first filtering Figure 2.1: Two-level Wavelet Pyramid Structure the low-frequency coefficients c 5 (see Figure 2.1) using the Laplacian filter and then use the measure of the sub-band energy for all wavelet coefficients which is defined as: e k = 1 MN M i=1 j=1 N c k (i, j) 2, k = 2, 3,... 4K, where M and N are the image dimensions of the coefficient c k and K is the number of decomposition levels (in this case K = 2) Spatial Envelope Besides color and texture, also the structure of an image consisting of the spatial properties of the scene, is also an important factor to be considered for image classification. In the work of Oliva and Torralba [6], they used a very low dimensional representation of the scene which is called spatial envelope for recognition of real world scenes. The spatial envelope captures the holistic property of an image. Hence, segmentation and processing of individual objects and regions does not need to be taken under consideration here anymore. The authors defined five different perceptual properties (Naturalness, Openness, Roughness, Expansion and Ruggedness) and designed features which can be used to characterize different scene categories. If an image has high degrees of Naturalness, Ruggedness and Expansion, it probably describes a natural landscape. Therefore, we can then consider the spatial envelope of an image as an important factor for indoor/outdoor classification Metadata EXIF data saved by the camera does not depend on the image content and provides information on how the images were taken. We could take advantages of such data to improve the performance of indoor/outdoor classification. Out of the variety of EXIF fields, only the tags which expose some condition with the indoor/outdoor class need to be considered for classification. It is worth noticing that not all images in our data set contain all of the considered data fields. Hence, we set the corresponding data to zero, when they are not present. 4

17 CHAPTER 2. BACKGROUND (a) Second Level DWT of Indoor Image (b) Second Level DWT of Outdoor Image Figure 2.2: Two-level Wavelet Decomposition 5

18 CHAPTER 2. BACKGROUND Sometimes when we want to take a picture of a person in a bar or a club, which is usually dark inside, we may want to fire the flash. Hence, the camera s flash is more often used, when taking an image inside as it is brighter outside. As shown in Table 2.3, we can see the percentages of flash firing in different scenes are different Indoor Outdoor 0.8 Percentage Flash off (0) or Flash on (1) Figure 2.3: Distribution of Flash Exposure time is defined as the time duration from opening to closing the shutter. It could be very large if we take a picture of a star orbit. It could also be very short when a clear picture of a fast moving object is going to be taken. The larger the value of exposure is, the longer the lights will be captured, which is suitable for the situation with bad illumination. However, small exposure time is perfect for a brighter scene. Figure 2.4 shows that images with exposure times of less than sec. are more likely to be of outdoor scenes. Aperture values illustrates the amount of the light that can be captured in a fixed amount of time. The bigger the aperture is, the more light can reach the sensor if the exposure time is constant. On the contrary, the smaller it is, the less will be absorbed. Artificial light is not as bright as sunlight. In this case, aperture values are considered to be useful because in general they get smaller when people take photos outside and bigger inside. See Figure 2.5. ISO speed rating describes the film speed, i.e. how much light the sensor needs to produce a picture. Higher ISO value allows people shooting pictures in a lower light or with small aperture value. But there would be some noise generated on the images. Thus, higher ISO value is suitable for the scene low of illumination which are likely to be indoor scenes. 2.2 Classification Classification is the problem of identifying the groups to which new individual observations belong. A classifier should be able to learn this mapping, based on existing samples, which are already labeled. In our work, we focus on binary classification. 6

19 CHAPTER 2. BACKGROUND P(ET indoor) P(ET outdoor) Figure 2.4: Distribution of Exposure Time [sec] P(AV indoor) P(AV outdoor) Figure 2.5: Distribution of Aperture Value K-nearest Neighbor (KNN) K-nearest neighbor algorithm is one of the simplest method in the field of machine learning. The idea of this classifier is that the training samples of the same category are grouped together in the multidimensional feature space. If most of the K nearest neighbors of a test sample belong to a single category, this test sample can also be included in this category. K can be determined by the user in the classification phase and usually it is an odd number which will not give any ties. As we can see in Figure 2.7 [1], the green circle can be 7

20 CHAPTER 2. BACKGROUND P(ISO indoor) P(ISO outdoor) Figure 2.6: Distribution of ISO Speed Rating classified as group of triangles if k = 3, since there are 2 triangles and only 1 square. If k = 5, the circle will then be classified as group of squares, since 3 squares win over 2 triangles. Figure 2.7: KNN Classification Support Vector Machine (SVM) Support vector machine is another very popular method in machine learning theory. A (p 1)-dimensional hyperplane is used to separate all p-dimensional data vectors into different groups. This is called a linear SVM. Assume we have some training data points D which have the form: 8 D = {(x i, y i ) x i R p, y i { 1, 1}} n i=1,

21 CHAPTER 2. BACKGROUND where x i is the p-dimensional vector and y i describes to which class the vector x i belongs. The hyperplane that separates the vectors can be expressed as: w x b = 0, where b is the bias value and w is normal vector of the hyperplane which can be calculated as: w = i α i y i x i where α is the Lagrange multiplier. There are a lot of hyperplanes available to be chosen. The best choice is the one whose distance or margin to the nearest data vector on each side is maximized, as we can see from the Figure 2.8 [2]. 1 w Figure 2.8: SVM Separating Hyperplane 2.3 Related Work Many researchers have worked on this interesting topic to detect whether an image was taken indoor or outdoor. Hence, many different techniques are proposed and evaluated. Most of them gave significant results by measuring different kinds of features. They extracted different image features by means of basic computer vision methods such as wavelet decomposition and used single feature or feature combinations to train classifiers. These classifiers were then used perform the final classification. First of all, color and texture are the most commonly used features. Szummer and Picard [10] used histograms of Ohta and RGB color space for extracting color features and it turns out that Ohta color space performs better than RGB, while in the work of Serrano et al. [9] the LST color space is used. Ohta and LST color spaces are almost the same, except for a scale factor. Serrano et al. [9] used 16 bin histograms instead of 32 bins in [10] to reduce the dimensionality by one half. On the other hand, the texture features are obtained from a two-level wavelet decomposition in [9] which gives better results than from multi-resolution simultaneous autoregressive model (MSAR) in [10]. 9

22 CHAPTER 2. BACKGROUND Secondly, edge analysis has also been used for indoor/outdoor classification. Payne and Singh [8] proposed that indoor images have more synthetic objects, therefore they have a greater proportion of edges that are straight compared to outdoor images. They used edge straightness as a feature and applied this method in a real-time system. Since most of the features which are directly extracted from image content are very extensively researched, exploiting EXIF metadata to classify indoor/outdoor images becomes a new perspective to improve the classification accuracy. In the work of Boutell and Luo [3] they took advantages of subject distance, exposure time, flash fired, aperture value and also combination of them. Moreover, they also added the metadata to content based cues which gives the highest accuracy. A number of classification techniques are exploited, such as K-Nearest Neighbors (KNN) [8] [10], support vector machine (SVM) [8] [9], neural network [11] and baysian network structure learning [4]. Szummer and Picard [10] used intersection norm for KNN instead of Euclidean norm for measuring the distance between histograms. But K-Nearest Neighbors is slow when the data sets are very large and feature dimension is high. Also, the optimal value of k is difficult to determine. However, SVM only needs to store the support vectors. Thus, its memory consumption is smaller than KNN method, but SVM has been criticized [5] for being too large to be used in a practical system with limited memory. 10

23 Chapter 3 Experiments and Results In our experiments, all low-level features (color, texture, spatial envelope) are extracted first and then we use every single feature as a classification criterion as in the work of Payne and Singh [7]. In a second batch of experiments, different feature combinations are used to train the classifier. Furthermore, among all the previous classification results the best combination is chosen to be combined with the metadata and this combination is used as a new classification criterion. Some other factors (different image division strategies and different resolutions) can be considered to affect the performance of the classification as well. Thus, a series of diverse experiments have been carried out in order to see how much impact these factors on the classification performance. 3.1 Image Data Set For the purpose of diversity, generalization and high accuracy, an image data set which consists of 5000 geotagged images from was exploited. The ground truth was manually labeled. There certainly exist images which we think their indoor-outdoor distinction is unclear, for example, a very close view of an object as shown in Figure 3.1(a) and Figure 3.1(b) or really ambiguous image as in Figure 3.1(c). Such images were omitted from the data set, which leaves us with 4525 images, in which 1395 images are indoor and 3130 images are outdoor. There is a class prior of 1 3 in our dataset. To let the result be more convincing, for evaluating the performance, we removed the class prior and the amounts of indoor and outdoor training images was balanced, i.e. 50% indoor images and 50% outdoor images. The scenes of the images range from indoor such as, living rooms, kitchens, concerts, clubs etc. to outdoor such as, sunset, lake, appearance of building, forest etc. 3.2 Experiment Protocol Before classification process begins, the whole data set is split into 3 parts: training set, validation set and test set. The test set is randomly selected with 300 indoor images and 300 outdoor images. The rest of the data set are for training and validation which contains 2190 images. The test set is totally independent from the training/validation set. If this would not be the case, it would lead to the situation that data from test set also contributes to the training process which could increase the accuracy implicitly. It is also worth noticing that the test, training and validation set were fixed for all experiments. Otherwise, the results would not be consistent and not comparable. Furthermore, some feature vectors are of higher order of magnitude and with higher dimension, such as color feature vectors. Some of them are of smaller one and with lower dimension, 11

24 C HAPTER 3. E XPERIMENTS AND R ESULTS (a) Example 1 (b) Example 2 (c) Example 3 Figure 3.1: Ambiguous Image Examples such as the texture feature vectors. It is useless if we concatenate them directly together, because the features with higher order of magnitude and higher dimension will dominate the performance. Therefore, we need to normalize feature vectors before combining them using the formula as follows 0 v = 12 v µv, σv

25 CHAPTER 3. EXPERIMENTS AND RESULTS where µ v is mean value of the old data and σ v is standard deviation of the old data. Memory consumption was measured by the size of the training data in the classification phase for the KNN classifier and the size of support vectors for the SVM classifier. 3.3 Experiments with Single Image Features We extracted color, texture, spatial envelope and camera meta data from images and used these features to train SVM and KNN classifiers. As shown in Figure 3.2, the dimensions of the images are various, the properties are not comparable when the images are of different scales. Therefore we need to rescale all the images to the same resolution and then compare the properties. In this experiment, we normalized all Number of Images Image Resolution Figure 3.2: Histogram of Image Resolutions images to maximal 512 pixels on the dimension and also divided the images into 4x4 subblocks for feature extraction. It would be interesting to be noted that the values of meta data are unique no matter how the images are rescaled and divided, i.e. values of meta data and image size are independent. We can see from Table 3.1, spatial envelope performed better than other low-level features for both classifiers. Unlike color or texture feature, spatial envelope described the holistic property of an image, it gave more accurate information of the images. The precision of outdoor scenes is better than indoor scenes for spatial envelope. This is because the properties (Naturalness, Openness, Roughness, Expansion and Ruggedness) described in the work of Oliva and Torralba [6] contribute more to outdoor scenes. The image which possess less degree of such properties would probably be labeled as indoor. Therefore, the outdoor scenes are better classified. Also, the metadata alone performed best among other low-level features, even better than spatial envelope. The accuracy is not so good as stated 91.51% in the work [3] on our dataset. This is because the number of images who contained meta data shown in Table 3.2 is not the same in both works and not every image in our data set contain all types of meta data. 13

26 CHAPTER 3. EXPERIMENTS AND RESULTS Feature Classifier Indoor Outdoor Accuracy Color KNN SVM Texture KNN SVM Spatial Envelope KNN SVM Metadata KNN SVM Table 3.1: Results of Single Image Features Meta data Flash Exposure Time Aperture Value ISO Speed Rating All 4 meta data Percentage 92.11% 91.87% 53.81% 76.66% 42.30% Table 3.2: Percentage of images which have Metadata 3.4 Experiments with Feature Combinations We may be also concerned about the performance of the feature combination. The results are shown in Feature Classifier Accuracy Color KNN % 8.83% 2. Texture SVM % 1.17% 1. Color KNN % 3.33% 2. Spatial Envelope SVM % -3.82% 1. Texture KNN % -0.34% 2. Spatial Envelope SVM % -0.32% 1. Color, 2. Texture KNN % 17.50% 4.00% 3. Spatial Envelope SVM % 6.00% 1.18% Table 3.3: Results of Feature Combinations the Table 3.3. i (i = 1,2,3) means the accuracy increment corresponding to the i-th component. When features were combined and the dimensionality increased, the performance would then be improved. Thus, color-texture performed better than single color or texture feature. Color-texture-spatial envelope performed better than color-texture, color-spatial envelope and texture-spatial envelope. We have also noticed that the performance of single texture feature was not very great. However, after combining with other low-level features, the performance was significantly improved. 14

27 3.5 Experiments Integrated with Metadata CHAPTER 3. EXPERIMENTS AND RESULTS The meta data alone had the best results among all other low-level features. We are also interested in the performance of low-level features after integrating with meta data. In general, the performance was improved for most features. But for spatial envelope with meta data when classified by KNN classifier, there is a slight reduction on the performance as shown in Table 3.4. For feature combinations as shown in Table 3.5, the Feature + Metadata Classifier Accuracy Color KNN % SVM % Texture KNN % SVM % Spatial Envelope KNN % SVM % Table 3.4: Results of Single Features with Metadata performance were also improved for most cases. But the combination of color and spatial envelope became the best combination when classified by KNN, although there was a slight accuracy reduction. For SVM classifier, color-texture-spatial envelope still performed best. Feature + Meta data Classifier Accuracy Color-Texture KNN % SVM % Color- KNN % Spatial Envelope SVM % Texture- KNN % Spatial Envelope SVM % Color-Texture- KNN % Spatial Envelope SVM % Table 3.5: Results of Feature Combinations with Metadata 3.6 Experiments with Different Image Dimensions Since the dimensions of the images are various, it is not sufficient that we normalized all images only to one particular size. The time consumption was increasing while the resolution of images increases, since the images got bigger, it took longer to read and extract features from images. Hence, we expected that we can train and evaluate the performance with a relatively low image dimension without much influence on the final performance. We then normalized all images to different resolutions according to the maximal value of height and width which are 128, 256, 512, 1024, 2048 to see if the results are significantly different due to different resolutions. For spatial envelope, the code the authors provided in [6] resized all images to 15

28 CHAPTER 3. EXPERIMENTS AND RESULTS Color Features Texture Features 0.72 KNN SVM 0.72 KNN SVM Accuracy 0.64 Accuracy Image resolution Image resolution Figure 3.3: Color Feature and Texture Feature resolution 256x256 and camera metadata are independent on image dimensions, therefore we only evaluated color and texture features. As we can see from Figure 3.3, the accuracies don not change very much while the image dimension decreases. Especially, for color feature there is a slight increase at resolution 256. Hence, it would be a benefit to reduce image to a smaller dimension. 3.7 Experiments with Different Division Strategies Different parts of images give different semantic information. The small square which contains the black Figure 3.4: 3x3 vs. 4x4 point shown in Figure 3.4 contributes to the upper part when we use 4x4 division strategy. However, if the 3x3 division strategy is exploited, this part contributes to the middle part. The information it contains would influence the final prediction of the classifiers according to different spatial positions. We have carried out experiments, which divided images into 3x3 subblocks, 4x4 subblocks, 3 columns and 3 rows individually. 16

29 CHAPTER 3. EXPERIMENTS AND RESULTS For each strategy, we compute the feature vector for the whole image by concatenating the feature vectors of each block. We would like to investigate how spatial positions could influence the high-level decision. Color Features Texture Features 0.72 KNN SVM 0.72 KNN SVM Accuracy 0.64 Accuracy by 3 4 by 4 column wise row wise Different Feature Extraction Strategies 3 by 3 4 by 4 column wise row wise Different Feature Extraction Strategies Figure 3.5: Different Feature Extraction Strategies As we can see from Figure 3.5, the 4x4 strategy gave the best results for color features when classified by KNN, while column wise gave the best when classified by SVM. For texture features, column wise strategy was the best for KNN classifier and 3x3 was the best for SVM. 3.8 Experiments with Two Stage Support Vector Machine Figure 3.6: Illustration of the two-stage classification approach A two stage support vector machine approach was also employed as a comparison to the work of Serrano et al. [9]. The first stage trains the SVM with image features and we compute the distance between feature vectors and separating hyperplane denoted by f(x), which can be calculated as, f(x) = w x + b, where x is the feature vector, b is the bias value. In the second stage, a new SVM will be trained using the distance values f(x) obtained from the first stage 17

30 CHAPTER 3. EXPERIMENTS AND RESULTS and it will produce final classification results. The whole procedure is illustrated in Figure 3.6. In our experiments, for each block we gathered the color and texture feature vectors of all images separately. In the training process, we trained the first SVM of each block and gained the f(x i ) values, where i indicates the block number. Then for each image, we summed f(x i ) together and used this sum f(x i ) as training i data for the second SVM. Normal SVM Two Stage SVM Accuracy Memory Consumption MB MB Time Consumption s 12154s Table 3.6: Comparison between Normal SVM and Two Stage SVM As we can see from Table 3.6, compared to the method which concatenated color texture vectors together, the two stage vector machine achieves a better performance but it occupied 3 times the memory and took 10 times longer compared to the normal SVM. Compared to the work of Serrano et.al in [9], the performance was not so good as theirs on our dataset. Since we used the same technique as used in [9], we could conclude that the images from our dataset are more difficult to be identified as indoor or outdoor. 18

31 Chapter 4 Conclusion After these step by step experiments, we have shown that we can gain the high-level image property (indoor/outdoor) by integrating camera meta data with low-level image properties color, texture and shape. Additionally, we also considered some other factors such as, subblock division strategy and different resolutions of images, which could play important roles in this classification problem. We have seen that for single features, the spatial envelope performed best among all low-level features and camera meta data performed even better than spatial envelope. The feature combination with color, texture and spatial envelope had better accuracy than other combinations. After integrated with meta data, in some cases the accuracies decreased and color spatial envelope outperformed over other features. When the resolutions of images increased, the classification accuracy did not have significant growing tendency. Therefore, for the purpose of efficiency, we could reduce images to a relatively small dimension without much influence on the accuracy. With the experiments with two level support vector machine, we found that the performance on our data set is not so good compared to other image data set. For future work, we may need to build a common data set for evaluation of different methods, otherwise we cannot compare the performance from different works. And we believe that if more ground truth for more diverse image scenes are provided and more advanced classifiers are used, better accuracy would be achieved. 19

32 CHAPTER 4. CONCLUSION 20

33 Bibliography [1] neighbor algorithm. [2] vector machine. [3] Matthew R. Boutell and Jiebo Luo. Photo classification by integrating image content and camera metadata. In ICPR (4), pages , [4] Michael J. Kane and Andreas E. Savakis. Bayesian network structure learning and inference in indoor vs. outdoor image classification. In ICPR (2), pages , [5] Lei Zhang Mingjing. Boosting image orientation detection with indoor vs. outdoor classification, [6] Aude Oliva and Antonio Torralba. Modeling the shape of the scene: A holistic representation of the spatial envelope. International Journal of Computer Vision, 42: , [7] Andrew Payne and Sameer Singh. A benchmark for indoor/outdoor scene classification. In ICAPR (2), pages , [8] Andrew Payne and Sameer Singh. Indoor vs. outdoor scene classification in digital photographs. Pattern Recognition, 38(10): , [9] Navid Serrano, Andreas E. Savakis, and Jiebo Luo. A computationally efficient approach to indoor/outdoor scene classification. In ICPR (4), pages 146, [10] Martin Szummer and Rosalind W. Picard. Indoor-outdoor image classification. In CAIVD, pages 42 51, [11] Li Tao, Yeong-Hwa Kim, and Yeong-Taeg Kim. An efficient neural network based indoor-outdoor scene classification algorithm. In Consumer Electronics (ICCE), 2010 Digest of Technical Papers International Conference on, pages , jan

Evaluation of texture features for image segmentation

Evaluation of texture features for image segmentation RIT Scholar Works Articles 9-14-2001 Evaluation of texture features for image segmentation Navid Serrano Jiebo Luo Andreas Savakis Follow this and additional works at: http://scholarworks.rit.edu/article

More information

A Scene Recognition Algorithm Based on Covariance Descriptor

A Scene Recognition Algorithm Based on Covariance Descriptor A Scene Recognition Algorithm Based on Covariance Descriptor Yinghui Ge Faculty of Information Science and Technology Ningbo University Ningbo, China gyhzd@tom.com Jianjun Yu Department of Computer Science

More information

Beyond Bags of Features

Beyond Bags of Features : for Recognizing Natural Scene Categories Matching and Modeling Seminar Instructed by Prof. Haim J. Wolfson School of Computer Science Tel Aviv University December 9 th, 2015

More information

Learning and Inferring Depth from Monocular Images. Jiyan Pan April 1, 2009

Learning and Inferring Depth from Monocular Images. Jiyan Pan April 1, 2009 Learning and Inferring Depth from Monocular Images Jiyan Pan April 1, 2009 Traditional ways of inferring depth Binocular disparity Structure from motion Defocus Given a single monocular image, how to infer

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

Correcting User Guided Image Segmentation

Correcting User Guided Image Segmentation Correcting User Guided Image Segmentation Garrett Bernstein (gsb29) Karen Ho (ksh33) Advanced Machine Learning: CS 6780 Abstract We tackle the problem of segmenting an image into planes given user input.

More information

Discriminative classifiers for image recognition

Discriminative classifiers for image recognition Discriminative classifiers for image recognition May 26 th, 2015 Yong Jae Lee UC Davis Outline Last time: window-based generic object detection basic pipeline face detection with boosting as case study

More information

AUTOMATIC IMAGE TAGGING BY USING IMAGE CONTENT ANALYSIS

AUTOMATIC IMAGE TAGGING BY USING IMAGE CONTENT ANALYSIS AUTOMATIC IMAGE TAGGING BY USING IMAGE CONTENT ANALYSIS Lailatul Qadri binti Zakaria a, Paul Lewis b, Wendy Hall c Pusat Pengajian Teknologi Maklumat, Fakulti Teknologi dan Sains Maklumat, Universiti Kebangsaan

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

SUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS

SUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS SUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS Cognitive Robotics Original: David G. Lowe, 004 Summary: Coen van Leeuwen, s1460919 Abstract: This article presents a method to extract

More information

Visual localization using global visual features and vanishing points

Visual localization using global visual features and vanishing points Visual localization using global visual features and vanishing points Olivier Saurer, Friedrich Fraundorfer, and Marc Pollefeys Computer Vision and Geometry Group, ETH Zürich, Switzerland {saurero,fraundorfer,marc.pollefeys}@inf.ethz.ch

More information

More Learning. Ensembles Bayes Rule Neural Nets K-means Clustering EM Clustering WEKA

More Learning. Ensembles Bayes Rule Neural Nets K-means Clustering EM Clustering WEKA More Learning Ensembles Bayes Rule Neural Nets K-means Clustering EM Clustering WEKA 1 Ensembles An ensemble is a set of classifiers whose combined results give the final decision. test feature vector

More information

ELL 788 Computational Perception & Cognition July November 2015

ELL 788 Computational Perception & Cognition July November 2015 ELL 788 Computational Perception & Cognition July November 2015 Module 6 Role of context in object detection Objects and cognition Ambiguous objects Unfavorable viewing condition Context helps in object

More information

Mobile Human Detection Systems based on Sliding Windows Approach-A Review

Mobile Human Detection Systems based on Sliding Windows Approach-A Review Mobile Human Detection Systems based on Sliding Windows Approach-A Review Seminar: Mobile Human detection systems Njieutcheu Tassi cedrique Rovile Department of Computer Engineering University of Heidelberg

More information

Detecting and Identifying Moving Objects in Real-Time

Detecting and Identifying Moving Objects in Real-Time Chapter 9 Detecting and Identifying Moving Objects in Real-Time For surveillance applications or for human-computer interaction, the automated real-time tracking of moving objects in images from a stationary

More information

CS 534: Computer Vision Segmentation and Perceptual Grouping

CS 534: Computer Vision Segmentation and Perceptual Grouping CS 534: Computer Vision Segmentation and Perceptual Grouping Ahmed Elgammal Dept of Computer Science CS 534 Segmentation - 1 Outlines Mid-level vision What is segmentation Perceptual Grouping Segmentation

More information

Character Recognition from Google Street View Images

Character Recognition from Google Street View Images Character Recognition from Google Street View Images Indian Institute of Technology Course Project Report CS365A By Ritesh Kumar (11602) and Srikant Singh (12729) Under the guidance of Professor Amitabha

More information

Artifacts and Textured Region Detection

Artifacts and Textured Region Detection Artifacts and Textured Region Detection 1 Vishal Bangard ECE 738 - Spring 2003 I. INTRODUCTION A lot of transformations, when applied to images, lead to the development of various artifacts in them. In

More information

Texture Analysis of Painted Strokes 1) Martin Lettner, Paul Kammerer, Robert Sablatnig

Texture Analysis of Painted Strokes 1) Martin Lettner, Paul Kammerer, Robert Sablatnig Texture Analysis of Painted Strokes 1) Martin Lettner, Paul Kammerer, Robert Sablatnig Vienna University of Technology, Institute of Computer Aided Automation, Pattern Recognition and Image Processing

More information

Learning to Recognize Faces in Realistic Conditions

Learning to Recognize Faces in Realistic Conditions 000 001 002 003 004 005 006 007 008 009 010 011 012 013 014 015 016 017 018 019 020 021 022 023 024 025 026 027 028 029 030 031 032 033 034 035 036 037 038 039 040 041 042 043 044 045 046 047 048 049 050

More information

Data-driven Depth Inference from a Single Still Image

Data-driven Depth Inference from a Single Still Image Data-driven Depth Inference from a Single Still Image Kyunghee Kim Computer Science Department Stanford University kyunghee.kim@stanford.edu Abstract Given an indoor image, how to recover its depth information

More information

Real-Time Detection of Landscape Scenes

Real-Time Detection of Landscape Scenes Real-Time Detection of Landscape Scenes Sami Huttunen 1,EsaRahtu 1, Iivari Kunttu 2, Juuso Gren 2, and Janne Heikkilä 1 1 Machine Vision Group, University of Oulu, Finland firstname.lastname@ee.oulu.fi

More information

Traffic Signs Recognition using HP and HOG Descriptors Combined to MLP and SVM Classifiers

Traffic Signs Recognition using HP and HOG Descriptors Combined to MLP and SVM Classifiers Traffic Signs Recognition using HP and HOG Descriptors Combined to MLP and SVM Classifiers A. Salhi, B. Minaoui, M. Fakir, H. Chakib, H. Grimech Faculty of science and Technology Sultan Moulay Slimane

More information

Using Machine Learning for Classification of Cancer Cells

Using Machine Learning for Classification of Cancer Cells Using Machine Learning for Classification of Cancer Cells Camille Biscarrat University of California, Berkeley I Introduction Cell screening is a commonly used technique in the development of new drugs.

More information

arxiv: v3 [cs.cv] 3 Oct 2012

arxiv: v3 [cs.cv] 3 Oct 2012 Combined Descriptors in Spatial Pyramid Domain for Image Classification Junlin Hu and Ping Guo arxiv:1210.0386v3 [cs.cv] 3 Oct 2012 Image Processing and Pattern Recognition Laboratory Beijing Normal University,

More information

Robotics Programming Laboratory

Robotics Programming Laboratory Chair of Software Engineering Robotics Programming Laboratory Bertrand Meyer Jiwon Shin Lecture 8: Robot Perception Perception http://pascallin.ecs.soton.ac.uk/challenges/voc/databases.html#caltech car

More information

Introduction to object recognition. Slides adapted from Fei-Fei Li, Rob Fergus, Antonio Torralba, and others

Introduction to object recognition. Slides adapted from Fei-Fei Li, Rob Fergus, Antonio Torralba, and others Introduction to object recognition Slides adapted from Fei-Fei Li, Rob Fergus, Antonio Torralba, and others Overview Basic recognition tasks A statistical learning approach Traditional or shallow recognition

More information

Query by Fax for Content-Based Image Retrieval

Query by Fax for Content-Based Image Retrieval Query by Fax for Content-Based Image Retrieval Mohammad F. A. Fauzi and Paul H. Lewis Intelligence, Agents and Multimedia Group, Department of Electronics and Computer Science, University of Southampton,

More information

Manual Exposure without a Meter

Manual Exposure without a Meter Manual Exposure without a Meter Scotty Elmslie, June 2018 Many books explain how to use your camera s meter the various metering modes, the difference between incident and reflective metering, how to compensate

More information

Using the Forest to See the Trees: Context-based Object Recognition

Using the Forest to See the Trees: Context-based Object Recognition Using the Forest to See the Trees: Context-based Object Recognition Bill Freeman Joint work with Antonio Torralba and Kevin Murphy Computer Science and Artificial Intelligence Laboratory MIT A computer

More information

CHAPTER 8 COMPOUND CHARACTER RECOGNITION USING VARIOUS MODELS

CHAPTER 8 COMPOUND CHARACTER RECOGNITION USING VARIOUS MODELS CHAPTER 8 COMPOUND CHARACTER RECOGNITION USING VARIOUS MODELS 8.1 Introduction The recognition systems developed so far were for simple characters comprising of consonants and vowels. But there is one

More information

Schedule for Rest of Semester

Schedule for Rest of Semester Schedule for Rest of Semester Date Lecture Topic 11/20 24 Texture 11/27 25 Review of Statistics & Linear Algebra, Eigenvectors 11/29 26 Eigenvector expansions, Pattern Recognition 12/4 27 Cameras & calibration

More information

Introduction to Shutter Speed in Digital Photography. Read more:

Introduction to Shutter Speed in Digital Photography. Read more: Introduction to Shutter Speed in Digital Photography Read more: http://digital-photography-school.com/shutterspeed#ixzz26mrybgum What is Shutter Speed? shutter speed is the amount of time that the shutter

More information

Tri-modal Human Body Segmentation

Tri-modal Human Body Segmentation Tri-modal Human Body Segmentation Master of Science Thesis Cristina Palmero Cantariño Advisor: Sergio Escalera Guerrero February 6, 2014 Outline 1 Introduction 2 Tri-modal dataset 3 Proposed baseline 4

More information

Chapter 4. The Classification of Species and Colors of Finished Wooden Parts Using RBFNs

Chapter 4. The Classification of Species and Colors of Finished Wooden Parts Using RBFNs Chapter 4. The Classification of Species and Colors of Finished Wooden Parts Using RBFNs 4.1 Introduction In Chapter 1, an introduction was given to the species and color classification problem of kitchen

More information

CHAPTER 3 SURFACE ROUGHNESS

CHAPTER 3 SURFACE ROUGHNESS 38 CHAPTER 3 SURFACE ROUGHNESS 3.1 SURFACE ROUGHNESS AND ITS IMPORTANCE The evaluation of surface roughness of machined parts using a direct contact method has limited flexibility in handling the different

More information

Machine Learning Classifiers and Boosting

Machine Learning Classifiers and Boosting Machine Learning Classifiers and Boosting Reading Ch 18.6-18.12, 20.1-20.3.2 Outline Different types of learning problems Different types of learning algorithms Supervised learning Decision trees Naïve

More information

FACULTY OF ENGINEERING AND INFORMATION TECHNOLOGY DEPARTMENT OF COMPUTER SCIENCE. Project Plan

FACULTY OF ENGINEERING AND INFORMATION TECHNOLOGY DEPARTMENT OF COMPUTER SCIENCE. Project Plan FACULTY OF ENGINEERING AND INFORMATION TECHNOLOGY DEPARTMENT OF COMPUTER SCIENCE Project Plan Structured Object Recognition for Content Based Image Retrieval Supervisors: Dr. Antonio Robles Kelly Dr. Jun

More information

A GENETIC ALGORITHM FOR MOTION DETECTION

A GENETIC ALGORITHM FOR MOTION DETECTION A GENETIC ALGORITHM FOR MOTION DETECTION Jarosław Mamica, Tomasz Walkowiak Institute of Engineering Cybernetics, Wrocław University of Technology ul. Janiszewskiego 11/17, 50-37 Wrocław, POLAND, Phone:

More information

Every Picture Tells a Story: Generating Sentences from Images

Every Picture Tells a Story: Generating Sentences from Images Every Picture Tells a Story: Generating Sentences from Images Ali Farhadi, Mohsen Hejrati, Mohammad Amin Sadeghi, Peter Young, Cyrus Rashtchian, Julia Hockenmaier, David Forsyth University of Illinois

More information

Semantic Visual Decomposition Modelling for Improving Object Detection in Complex Scene Images

Semantic Visual Decomposition Modelling for Improving Object Detection in Complex Scene Images Semantic Visual Decomposition Modelling for Improving Object Detection in Complex Scene Images Ge Qin Department of Computing University of Surrey United Kingdom g.qin@surrey.ac.uk Bogdan Vrusias Department

More information

Automatic Colorization of Grayscale Images

Automatic Colorization of Grayscale Images Automatic Colorization of Grayscale Images Austin Sousa Rasoul Kabirzadeh Patrick Blaes Department of Electrical Engineering, Stanford University 1 Introduction ere exists a wealth of photographic images,

More information

Feature Extractors. CS 188: Artificial Intelligence Fall Some (Vague) Biology. The Binary Perceptron. Binary Decision Rule.

Feature Extractors. CS 188: Artificial Intelligence Fall Some (Vague) Biology. The Binary Perceptron. Binary Decision Rule. CS 188: Artificial Intelligence Fall 2008 Lecture 24: Perceptrons II 11/24/2008 Dan Klein UC Berkeley Feature Extractors A feature extractor maps inputs to feature vectors Dear Sir. First, I must solicit

More information

An Empirical Study of Lazy Multilabel Classification Algorithms

An Empirical Study of Lazy Multilabel Classification Algorithms An Empirical Study of Lazy Multilabel Classification Algorithms E. Spyromitros and G. Tsoumakas and I. Vlahavas Department of Informatics, Aristotle University of Thessaloniki, 54124 Thessaloniki, Greece

More information

Adaptive Gesture Recognition System Integrating Multiple Inputs

Adaptive Gesture Recognition System Integrating Multiple Inputs Adaptive Gesture Recognition System Integrating Multiple Inputs Master Thesis - Colloquium Tobias Staron University of Hamburg Faculty of Mathematics, Informatics and Natural Sciences Technical Aspects

More information

Fast Fuzzy Clustering of Infrared Images. 2. brfcm

Fast Fuzzy Clustering of Infrared Images. 2. brfcm Fast Fuzzy Clustering of Infrared Images Steven Eschrich, Jingwei Ke, Lawrence O. Hall and Dmitry B. Goldgof Department of Computer Science and Engineering, ENB 118 University of South Florida 4202 E.

More information

Spam Filtering Using Visual Features

Spam Filtering Using Visual Features Spam Filtering Using Visual Features Sirnam Swetha Computer Science Engineering sirnam.swetha@research.iiit.ac.in Sharvani Chandu Electronics and Communication Engineering sharvani.chandu@students.iiit.ac.in

More information

Automatic Image Orientation Determination with Natural Image Statistics

Automatic Image Orientation Determination with Natural Image Statistics TR2005-545, October 2004, Department of Computer Science, Dartmouth College Automatic Image Orientation Determination with Natural Image Statistics Siwei Lyu Department of Computer Science Dartmouth College

More information

Contexts and 3D Scenes

Contexts and 3D Scenes Contexts and 3D Scenes Computer Vision Jia-Bin Huang, Virginia Tech Many slides from D. Hoiem Administrative stuffs Final project presentation Nov 30 th 3:30 PM 4:45 PM Grading Three senior graders (30%)

More information

Applying Supervised Learning

Applying Supervised Learning Applying Supervised Learning When to Consider Supervised Learning A supervised learning algorithm takes a known set of input data (the training set) and known responses to the data (output), and trains

More information

Predicting Messaging Response Time in a Long Distance Relationship

Predicting Messaging Response Time in a Long Distance Relationship Predicting Messaging Response Time in a Long Distance Relationship Meng-Chen Shieh m3shieh@ucsd.edu I. Introduction The key to any successful relationship is communication, especially during times when

More information

More on Learning. Neural Nets Support Vectors Machines Unsupervised Learning (Clustering) K-Means Expectation-Maximization

More on Learning. Neural Nets Support Vectors Machines Unsupervised Learning (Clustering) K-Means Expectation-Maximization More on Learning Neural Nets Support Vectors Machines Unsupervised Learning (Clustering) K-Means Expectation-Maximization Neural Net Learning Motivated by studies of the brain. A network of artificial

More information

FACE DETECTION AND RECOGNITION OF DRAWN CHARACTERS HERMAN CHAU

FACE DETECTION AND RECOGNITION OF DRAWN CHARACTERS HERMAN CHAU FACE DETECTION AND RECOGNITION OF DRAWN CHARACTERS HERMAN CHAU 1. Introduction Face detection of human beings has garnered a lot of interest and research in recent years. There are quite a few relatively

More information

2. LITERATURE REVIEW

2. LITERATURE REVIEW 2. LITERATURE REVIEW CBIR has come long way before 1990 and very little papers have been published at that time, however the number of papers published since 1997 is increasing. There are many CBIR algorithms

More information

Indoor Object Recognition of 3D Kinect Dataset with RNNs

Indoor Object Recognition of 3D Kinect Dataset with RNNs Indoor Object Recognition of 3D Kinect Dataset with RNNs Thiraphat Charoensripongsa, Yue Chen, Brian Cheng 1. Introduction Recent work at Stanford in the area of scene understanding has involved using

More information

Dynamic Routing Between Capsules

Dynamic Routing Between Capsules Report Explainable Machine Learning Dynamic Routing Between Capsules Author: Michael Dorkenwald Supervisor: Dr. Ullrich Köthe 28. Juni 2018 Inhaltsverzeichnis 1 Introduction 2 2 Motivation 2 3 CapusleNet

More information

Report: Privacy-Preserving Classification on Deep Neural Network

Report: Privacy-Preserving Classification on Deep Neural Network Report: Privacy-Preserving Classification on Deep Neural Network Janno Veeorg Supervised by Helger Lipmaa and Raul Vicente Zafra May 25, 2017 1 Introduction In this report we consider following task: how

More information

Understanding Faces. Detection, Recognition, and. Transformation of Faces 12/5/17

Understanding Faces. Detection, Recognition, and. Transformation of Faces 12/5/17 Understanding Faces Detection, Recognition, and 12/5/17 Transformation of Faces Lucas by Chuck Close Chuck Close, self portrait Some slides from Amin Sadeghi, Lana Lazebnik, Silvio Savarese, Fei-Fei Li

More information

Instantaneously trained neural networks with complex inputs

Instantaneously trained neural networks with complex inputs Louisiana State University LSU Digital Commons LSU Master's Theses Graduate School 2003 Instantaneously trained neural networks with complex inputs Pritam Rajagopal Louisiana State University and Agricultural

More information

Business Club. Decision Trees

Business Club. Decision Trees Business Club Decision Trees Business Club Analytics Team December 2017 Index 1. Motivation- A Case Study 2. The Trees a. What is a decision tree b. Representation 3. Regression v/s Classification 4. Building

More information

Improved Spatial Pyramid Matching for Image Classification

Improved Spatial Pyramid Matching for Image Classification Improved Spatial Pyramid Matching for Image Classification Mohammad Shahiduzzaman, Dengsheng Zhang, and Guojun Lu Gippsland School of IT, Monash University, Australia {Shahid.Zaman,Dengsheng.Zhang,Guojun.Lu}@monash.edu

More information

Online structured learning for Obstacle avoidance

Online structured learning for Obstacle avoidance Adarsh Kowdle Cornell University Zhaoyin Jia Cornell University apk64@cornell.edu zj32@cornell.edu Abstract Obstacle avoidance based on a monocular system has become a very interesting area in robotics

More information

Disguised Face Identification (DFI) with Facial KeyPoints using Spatial Fusion Convolutional Network. Nathan Sun CIS601

Disguised Face Identification (DFI) with Facial KeyPoints using Spatial Fusion Convolutional Network. Nathan Sun CIS601 Disguised Face Identification (DFI) with Facial KeyPoints using Spatial Fusion Convolutional Network Nathan Sun CIS601 Introduction Face ID is complicated by alterations to an individual s appearance Beard,

More information

CHAPTER 9. Classification Scheme Using Modified Photometric. Stereo and 2D Spectra Comparison

CHAPTER 9. Classification Scheme Using Modified Photometric. Stereo and 2D Spectra Comparison CHAPTER 9 Classification Scheme Using Modified Photometric Stereo and 2D Spectra Comparison 9.1. Introduction In Chapter 8, even we combine more feature spaces and more feature generators, we note that

More information

Naïve Bayes for text classification

Naïve Bayes for text classification Road Map Basic concepts Decision tree induction Evaluation of classifiers Rule induction Classification using association rules Naïve Bayesian classification Naïve Bayes for text classification Support

More information

Classification and Detection in Images. D.A. Forsyth

Classification and Detection in Images. D.A. Forsyth Classification and Detection in Images D.A. Forsyth Classifying Images Motivating problems detecting explicit images classifying materials classifying scenes Strategy build appropriate image features train

More information

Scene Recognition using Bag-of-Words

Scene Recognition using Bag-of-Words Scene Recognition using Bag-of-Words Sarthak Ahuja B.Tech Computer Science Indraprastha Institute of Information Technology Okhla, Delhi 110020 Email: sarthak12088@iiitd.ac.in Anchita Goel B.Tech Computer

More information

Discovering Visual Hierarchy through Unsupervised Learning Haider Razvi

Discovering Visual Hierarchy through Unsupervised Learning Haider Razvi Discovering Visual Hierarchy through Unsupervised Learning Haider Razvi hrazvi@stanford.edu 1 Introduction: We present a method for discovering visual hierarchy in a set of images. Automatically grouping

More information

CS4442/9542b Artificial Intelligence II prof. Olga Veksler

CS4442/9542b Artificial Intelligence II prof. Olga Veksler CS4442/9542b Artificial Intelligence II prof. Olga Veksler Lecture 2 Computer Vision Introduction, Filtering Some slides from: D. Jacobs, D. Lowe, S. Seitz, A.Efros, X. Li, R. Fergus, J. Hayes, S. Lazebnik,

More information

Storyline Reconstruction for Unordered Images

Storyline Reconstruction for Unordered Images Introduction: Storyline Reconstruction for Unordered Images Final Paper Sameedha Bairagi, Arpit Khandelwal, Venkatesh Raizaday Storyline reconstruction is a relatively new topic and has not been researched

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

Leaf Image Recognition Based on Wavelet and Fractal Dimension

Leaf Image Recognition Based on Wavelet and Fractal Dimension Journal of Computational Information Systems 11: 1 (2015) 141 148 Available at http://www.jofcis.com Leaf Image Recognition Based on Wavelet and Fractal Dimension Haiyan ZHANG, Xingke TAO School of Information,

More information

Using Machine Learning to Optimize Storage Systems

Using Machine Learning to Optimize Storage Systems Using Machine Learning to Optimize Storage Systems Dr. Kiran Gunnam 1 Outline 1. Overview 2. Building Flash Models using Logistic Regression. 3. Storage Object classification 4. Storage Allocation recommendation

More information

Automatic Classification of Outdoor Images by Region Matching

Automatic Classification of Outdoor Images by Region Matching Automatic Classification of Outdoor Images by Region Matching Oliver van Kaick and Greg Mori School of Computing Science Simon Fraser University, Burnaby, BC, V5A S6 Canada E-mail: {ovankaic,mori}@cs.sfu.ca

More information

Homework. Gaussian, Bishop 2.3 Non-parametric, Bishop 2.5 Linear regression Pod-cast lecture on-line. Next lectures:

Homework. Gaussian, Bishop 2.3 Non-parametric, Bishop 2.5 Linear regression Pod-cast lecture on-line. Next lectures: Homework Gaussian, Bishop 2.3 Non-parametric, Bishop 2.5 Linear regression 3.0-3.2 Pod-cast lecture on-line Next lectures: I posted a rough plan. It is flexible though so please come with suggestions Bayes

More information

Lab 9. Julia Janicki. Introduction

Lab 9. Julia Janicki. Introduction Lab 9 Julia Janicki Introduction My goal for this project is to map a general land cover in the area of Alexandria in Egypt using supervised classification, specifically the Maximum Likelihood and Support

More information

Ensemble of Bayesian Filters for Loop Closure Detection

Ensemble of Bayesian Filters for Loop Closure Detection Ensemble of Bayesian Filters for Loop Closure Detection Mohammad Omar Salameh, Azizi Abdullah, Shahnorbanun Sahran Pattern Recognition Research Group Center for Artificial Intelligence Faculty of Information

More information

Analysis: TextonBoost and Semantic Texton Forests. Daniel Munoz Februrary 9, 2009

Analysis: TextonBoost and Semantic Texton Forests. Daniel Munoz Februrary 9, 2009 Analysis: TextonBoost and Semantic Texton Forests Daniel Munoz 16-721 Februrary 9, 2009 Papers [shotton-eccv-06] J. Shotton, J. Winn, C. Rother, A. Criminisi, TextonBoost: Joint Appearance, Shape and Context

More information

Kapitel 4: Clustering

Kapitel 4: Clustering Ludwig-Maximilians-Universität München Institut für Informatik Lehr- und Forschungseinheit für Datenbanksysteme Knowledge Discovery in Databases WiSe 2017/18 Kapitel 4: Clustering Vorlesung: Prof. Dr.

More information

Image Analysis. 1. A First Look at Image Classification

Image Analysis. 1. A First Look at Image Classification Image Analysis Image Analysis 1. A First Look at Image Classification Lars Schmidt-Thieme Information Systems and Machine Learning Lab (ISMLL) Institute for Business Economics and Information Systems &

More information

Feature Extractors. CS 188: Artificial Intelligence Fall Nearest-Neighbor Classification. The Perceptron Update Rule.

Feature Extractors. CS 188: Artificial Intelligence Fall Nearest-Neighbor Classification. The Perceptron Update Rule. CS 188: Artificial Intelligence Fall 2007 Lecture 26: Kernels 11/29/2007 Dan Klein UC Berkeley Feature Extractors A feature extractor maps inputs to feature vectors Dear Sir. First, I must solicit your

More information

FEATURE EXTRACTION TECHNIQUES FOR IMAGE RETRIEVAL USING HAAR AND GLCM

FEATURE EXTRACTION TECHNIQUES FOR IMAGE RETRIEVAL USING HAAR AND GLCM FEATURE EXTRACTION TECHNIQUES FOR IMAGE RETRIEVAL USING HAAR AND GLCM Neha 1, Tanvi Jain 2 1,2 Senior Research Fellow (SRF), SAM-C, Defence R & D Organization, (India) ABSTRACT Content Based Image Retrieval

More information

Local Features: Detection, Description & Matching

Local Features: Detection, Description & Matching Local Features: Detection, Description & Matching Lecture 08 Computer Vision Material Citations Dr George Stockman Professor Emeritus, Michigan State University Dr David Lowe Professor, University of British

More information

Texture. Frequency Descriptors. Frequency Descriptors. Frequency Descriptors. Frequency Descriptors. Frequency Descriptors

Texture. Frequency Descriptors. Frequency Descriptors. Frequency Descriptors. Frequency Descriptors. Frequency Descriptors Texture The most fundamental question is: How can we measure texture, i.e., how can we quantitatively distinguish between different textures? Of course it is not enough to look at the intensity of individual

More information

CPSC 340: Machine Learning and Data Mining

CPSC 340: Machine Learning and Data Mining CPSC 340: Machine Learning and Data Mining Fundamentals of learning (continued) and the k-nearest neighbours classifier Original version of these slides by Mark Schmidt, with modifications by Mike Gelbart.

More information

Object Recognition Using Pictorial Structures. Daniel Huttenlocher Computer Science Department. In This Talk. Object recognition in computer vision

Object Recognition Using Pictorial Structures. Daniel Huttenlocher Computer Science Department. In This Talk. Object recognition in computer vision Object Recognition Using Pictorial Structures Daniel Huttenlocher Computer Science Department Joint work with Pedro Felzenszwalb, MIT AI Lab In This Talk Object recognition in computer vision Brief definition

More information

CS4442/9542b Artificial Intelligence II prof. Olga Veksler

CS4442/9542b Artificial Intelligence II prof. Olga Veksler CS4442/9542b Artificial Intelligence II prof. Olga Veksler Lecture 8 Computer Vision Introduction, Filtering Some slides from: D. Jacobs, D. Lowe, S. Seitz, A.Efros, X. Li, R. Fergus, J. Hayes, S. Lazebnik,

More information

Sketchable Histograms of Oriented Gradients for Object Detection

Sketchable Histograms of Oriented Gradients for Object Detection Sketchable Histograms of Oriented Gradients for Object Detection No Author Given No Institute Given Abstract. In this paper we investigate a new representation approach for visual object recognition. The

More information

Linear Regression and K-Nearest Neighbors 3/28/18

Linear Regression and K-Nearest Neighbors 3/28/18 Linear Regression and K-Nearest Neighbors 3/28/18 Linear Regression Hypothesis Space Supervised learning For every input in the data set, we know the output Regression Outputs are continuous A number,

More information

Storage Efficient NL-Means Burst Denoising for Programmable Cameras

Storage Efficient NL-Means Burst Denoising for Programmable Cameras Storage Efficient NL-Means Burst Denoising for Programmable Cameras Brendan Duncan Stanford University brendand@stanford.edu Miroslav Kukla Stanford University mkukla@stanford.edu Abstract An effective

More information

Feature Descriptors. CS 510 Lecture #21 April 29 th, 2013

Feature Descriptors. CS 510 Lecture #21 April 29 th, 2013 Feature Descriptors CS 510 Lecture #21 April 29 th, 2013 Programming Assignment #4 Due two weeks from today Any questions? How is it going? Where are we? We have two umbrella schemes for object recognition

More information

COLOR TEXTURE CLASSIFICATION USING LOCAL & GLOBAL METHOD FEATURE EXTRACTION

COLOR TEXTURE CLASSIFICATION USING LOCAL & GLOBAL METHOD FEATURE EXTRACTION COLOR TEXTURE CLASSIFICATION USING LOCAL & GLOBAL METHOD FEATURE EXTRACTION 1 Subodh S.Bhoite, 2 Prof.Sanjay S.Pawar, 3 Mandar D. Sontakke, 4 Ajay M. Pol 1,2,3,4 Electronics &Telecommunication Engineering,

More information

(and what the numbers mean)

(and what the numbers mean) Using Neutral Density Filters (and what the numbers mean) What are ND filters Neutral grey filters that effectively reduce the amount of light entering the lens. On solid ND filters the light-stopping

More information

CS6716 Pattern Recognition

CS6716 Pattern Recognition CS6716 Pattern Recognition Prototype Methods Aaron Bobick School of Interactive Computing Administrivia Problem 2b was extended to March 25. Done? PS3 will be out this real soon (tonight) due April 10.

More information

Introduction to Machine Learning. Xiaojin Zhu

Introduction to Machine Learning. Xiaojin Zhu Introduction to Machine Learning Xiaojin Zhu jerryzhu@cs.wisc.edu Read Chapter 1 of this book: Xiaojin Zhu and Andrew B. Goldberg. Introduction to Semi- Supervised Learning. http://www.morganclaypool.com/doi/abs/10.2200/s00196ed1v01y200906aim006

More information

A New Algorithm for Shape Detection

A New Algorithm for Shape Detection IOSR Journal of Computer Engineering (IOSR-JCE) e-issn: 2278-0661,p-ISSN: 2278-8727, Volume 19, Issue 3, Ver. I (May.-June. 2017), PP 71-76 www.iosrjournals.org A New Algorithm for Shape Detection Hewa

More information

Face Recognition using Eigenfaces SMAI Course Project

Face Recognition using Eigenfaces SMAI Course Project Face Recognition using Eigenfaces SMAI Course Project Satarupa Guha IIIT Hyderabad 201307566 satarupa.guha@research.iiit.ac.in Ayushi Dalmia IIIT Hyderabad 201307565 ayushi.dalmia@research.iiit.ac.in Abstract

More information

Contexts and 3D Scenes

Contexts and 3D Scenes Contexts and 3D Scenes Computer Vision Jia-Bin Huang, Virginia Tech Many slides from D. Hoiem Administrative stuffs Final project presentation Dec 1 st 3:30 PM 4:45 PM Goodwin Hall Atrium Grading Three

More information

Segmentation of Images

Segmentation of Images Segmentation of Images SEGMENTATION If an image has been preprocessed appropriately to remove noise and artifacts, segmentation is often the key step in interpreting the image. Image segmentation is a

More information

Classification of Digital Photos Taken by Photographers or Home Users

Classification of Digital Photos Taken by Photographers or Home Users Classification of Digital Photos Taken by Photographers or Home Users Hanghang Tong 1, Mingjing Li 2, Hong-Jiang Zhang 2, Jingrui He 1, and Changshui Zhang 3 1 Automation Department, Tsinghua University,

More information