Super-resolution Fusion using Adaptive Normalized Averaging
|
|
- Susan Cox
- 5 years ago
- Views:
Transcription
1 Super-resolution Fusion using Adaptive Normalized Averaging Tuan Q. Pham Lucas J. van Vliet Quantitative Imaging Group, Faculty of Applied Sciences, Delft University of Technology, Lorentzweg 1, 2628 CJ Delft, The Netherlands {tuan, lucas Keywords: super-resolution, sampling density, adaptive normalized averaging, irregular samples Abstract A fast method for super-resolution (SR) reconstruction from low resolution (LR) frames with known registration is proposed. The irregular LR samples are incorporated into the SR grid by stamping into 4-nearest neighbors with position certainties. The signal certainty reflects the errors in the LR pixels positions (computed by cross-correlation or optic flow) and their intensities. Adaptive normalized averaging is used in the fusion stage to enhance local linear structure and minimize further blurring. The local structure descriptors including orientation, anisotropy and curvature are computed directly on the SR grid and used as steering parameters for the fusion. The optimum scale for local fusion is achieved by a sample density transform, which is also presented for the first time in this paper. 1 Introduction Super-resolution from a sequence of low resolution images is an important step in early vision to increase spatial resolution of captured images for later recognition tasks. An extensive literature on this topic exists [1] [2], of which there are two main approaches: one with simultaneous estimation of image registration parameters and the high resolution image [3] [4] and the other with registration being computed separately before fusion [5] [6]. The super-resolution algorithm presented in this paper follows the second approach. In other words, given an estimate of the registration, our method attempts to reconstruct a high resolution image that is visually more informative than each of the individual low resolution frames. The main tasks involved in the fusion process are first to map the irregular pixels onto a rectangularly sampled SR grid and secondly to derive the SR grid values from them. The first attempt of such was initiated by Tsai and Huang [7], who assumed the aliased and noise-free low resolution images were originally sampled from a band-limited high resolution image. Kim, Bose and Valenzuela [5] extended this work to handle noisy and blurred LR images using recursive least squares optimization in the Fourier domain. A spatial domain method was then reported by Ur and Gross [8]. However, all these early algorithms only works with shifted LR inputs that can be aligned periodically within the SR grid. In a recent paper [6], Lertrattanapachich and Bose presented a spatial domain method based on Delaunay triangulation [9] that allows any arbitrary organization of input samples. Our structure adaptive fusion algorithm in this paper also can deal with irregular registration after a process of stamping pixels back onto the grid. However, it is much less sensitive to noise than a perfect surface reconstruction algorithm like [6] because of an averaging along the dominant local structure. The paper is organized as follows: section 2 briefly describes the signal/certainty principle in image processing via normalized averaging. Fusion of irregular and uncertain samples is possible under this principle by using 4-nearest neighbor stamping of the signal certainty. Section 4 introduces the density transform to compute sampling density of a set of irregular pixels. Section 5 combines adaptive filtering and density transform into the normalized averaging equation to achieve a structure-enhanced fusion result. Simulations and analysis of the super-fusion technique are given in section 6. Section 7 finally concludes the paper.
2 2 Normalized Averaging Normalized Averaging (NA), a special case of Normalized Convolution [10], is a local signal modelling method that takes signal uncertainty into account. The certainty of the signal may reflect noise found in the image as well as in the registration. Using a weighted average with signal certainty contributing to the weights, the method aims at maximizing the reconstructed signal continuity while maintaining the least deviation from the unknown underlying signal in a least squares sense. This characteristic is desirable for various interpolation tasks such as image reconstruction from uncertain data, interpolation and super-resolution. In terms of efficiency, a normalized averaging operation requires only two convolutions (*) with a fixed applicability function a (see equation 1, where s is the input signal, c is the associated certainty and r is the reconstructed signal). This applicability function decides how the neighboring signals influence the local signal model. Generally speaking, any local function that decays outside a certain radius can be used as the applicability function. Amongst those, a Gaussian function is the most popular due to its good localization in both spatial and frequency domain and a fast implementation [11]. r = (s.c) a c a (1) One simple illustration of the signal/certainty principle and normalized averaging is to reconstruct an image from sparse samples. In this example, a signal certainty of one is assigned to the available pixels and zero to the unavailable ones. Note that a simple Gaussian convolution of the degraded signal convey no information; whereas a normalized averaging operation results in a desirable outcome (figure 1). This interpolation result is comparable to a surface reconstruction technique like the Delaunay triangulation method [9] (built-in MATLAB function gridata), but it is much more efficient and robust to noise. Figure 1: (a) input with only 10% samples, (b) convolution result with a Gaussian kernel σ = 1, (c) normalized averaging result using a Gaussian σ = 1, (d) linear interpolation by Delaunay triangulation. 3 Fusion of Irregular Samples using 4- NN Stamping Image fusion of irregularly sampled data is also possible using normalized averaging approach. Difficulties arise, however, because the convolution with a Gaussian applicability function has to be performed on signal values registered at off-grid positions on the SR grid. In a recent paper, Andersson and Knutsson [12] proposed continuous normalized convolution to approximate the Gaussian spreading of an off-grid sample by truncating the local Taylor expansion up to the first derivative order. The approximation, though fast, does not perform well for small applicability function because of inaccuracy in derivative estimation. Pham and van Vliet [13], in such cases, lookup the filter coefficients from a table and do a direct accumulation of the sample spreading on the SR grid. This method, though more accurate, would induce a lot of computations since each local kernel must be looked up separately. In this paper, we avoid convolution of irregular samples by first stamping those samples with certainties onto the reference SR grid, followed by normalized averaging on the SR grid. The stamping step, which partially assigns an off-grid sample to its four nearest on-grid neighbors (4-NN), divides the certainty into four parts in a bilinear fashion. The stamped signal, weighted with its corresponding certainty share, is accumulated to the weighted sum of signals stored at each pixel (sc accum ). The certainty share is also accumulated to a SR certainty image (c accum ). After every sample has been stamped, normalized averaging takes place on the SR grid as normal with the weighted sum of signals and the SR certainty. This whole process is illustrated in Figure 2. (1-p)(1-q).{s.c,c} q 1-q (1-p)q.{s.c,c} p : {s,c} : p(1-q).{s.c,c} 1-p pq.{s.c,c} SC accum sc a ulated r = c a } C accum ulated Figure 2: 4-NN stamping of an off-grid pixel (black point) onto the SR grid (white points), followed by normalized averaging of the accumulated signal and certainty. The accumulation of signal and certainty after the 4-NN stamping is actually a normalized averaging in nature. The bilinear weight distribution could be seen as the result of filtering the irregular sampled input with a bilinear filter. The base width of this filter is twice the pitch of the SR grid and the gray volume under the filter is one. Sampling the filtered input by the SR grid yields a space-variant filtering of the irregular input sample that only affects its 4 nearest neighbors. The smoothing factor of this filter varies but stays below a fraction of the SR grid size. In the worst case scenario when the off-grid pixel is registered halfway in between the SR grid, the space-variant low-pass filter then approximates a 2x2 uniform filter. Due to the r
3 associativity of convolution, the result of normalized averaging on the accumulated signal and certainty is equivalent to normalized averaging on the irregular data with a slightly bigger smoothing sigma. This extra smoothing is however negligible in our superresolution task compared to the blur caused by the optics and the near 100% fill-factor of CCD sensors. 4 Density Transform From the illustration in section 2, it is clear that normalized averaging is an effective method for signal interpolation. However, the quality of the reconstruction depends heavily on the scale of the applicability function. For example, a Gaussian applicability function with a small sigma yields a result similar to nearest neighbor interpolation with undesirable flatpatches effects. A larger Gaussian sigma, on the other hand, incurs excessive blurring to the output, which smooths out the wanted details. More importantly, since the density of the registered samples varies from point to point over the whole SR image, a single scale might introduce unnecessary smoothing over some region while it might not be large enough to cover large gaps in the data. In this section, a method for computing sample density is established. The result could then be used to control the automatic scale selection in the normalized averaging process. The density transform of a gray-scale certainty image is defined as the smallest radius of a pillbox, centered at each pixel position, that encompasses a total certainty value of at least one. The density transform of a SR certainty image (defined in section 3) gives an indication of how dense the SR image is sampled in the vicinity of each pixel. The scale selection for the image fusion process is based on this information to limit the number of contributions from the neighborhood in the normalized averaging process. This ensures a minimum degree of blurring everywhere in the reconstructed image. The density transform is estimated by interpolating responses to a filter bank of increasing radii. By definition, convolution of local image with a pillbox filter of radius equal to the density transform results in a response of one. Pillbox filters with smaller radii thus give responses less than one, and vice versa. The response function therefore increases monotonically with the filter radius. In fact, it could be shown that the response stays zero until the radius reaches a certain value (the distance transform [14]); then, from that point on it keeps increasing with a roughly parabolic shape. A good approximation of the density transform could therefore be computed by piecewise quadratic interpolation. To reduce quantization errors in the pillbox filter, we use a smooth pillbox model based on the cosine function with r 0 being the pillbox radius (equation 2). This smoothing model implies a minimum pillbox radius of 0.5. h pillbox (r ; r 0 ) = 1, r r 0 0.5; 0, r r ; (1 + cos(π(r r )))/2, otherwise. (2) Apparently, the number of radius samples and the spacings between them determine the accuracy of the algorithm. Depending on the need of the problem, these settings are chosen accordingly. For example, if many LR images are used to construct a SR image, the SR certainty image might be quite densely filled after the 4-NN stamping; accuracy is therefore desired for small density. A sequence with an exponentially increasing step-size such as r 0 = n/2 (n = 0, 1, 2,...) could be used in such situation. 5 Adaptive Normalized Averaging If in section 4 we compute a variable scale for the applicability function; in this section, we improve normalized averaging further by applying structure adaptive filtering to the equation. It is clear that LR images often suffer from severe aliasing that causes jaggedness along edges (figure??b). Normal interpolation algorithms, including the original NA in section 2, reduce these jaggedness by smoothing the edges in all directions, thus also smearing out important details. To avoid such over-smoothing, adaptive filters that are elongated along local orientation are used instead of an isotropic Gaussian kernel in the original NA equation (equation 1). The steering parameters for the adaptive filter is computed from the Gradient Structure Tensor (GST) as follows. 5.1 GST from irregular data Being a least squares estimate of the local gradient vectors, the GST (equation 3) is a reliable orientation estimator. Apart from the orientation (φ), which is the direct result of a principal component analysis on the GST, other important local descriptors such as anisotropy (A) and curvature (κ) can also be derived (equation 4). Anisotropy indicates how strong the edge or line is in the local neighborhood. Curvature is defined as the rate of change of orientation along the dominant linear structure. Note that due to a sudden jump at ±π/2, φ is not directly differentiable. However, we can use the double angle representation in a complex orientation mapping to assure continuity for the differiation [15] [16]. T = I I T = λ u uu T + λ v vv T (3) φ = arg(u) A = (λ u λ v )/(λ u + λ v ) (4)
4 y = 1 2 u ϕ 2 κ x v Figure 3: (a) curvature-bent scale-adaptive anisotropic Gaussian kernel, (b) variable density input with adaptive filters, (c) fusion result with missing holes properly filled in. κ = φ/ v = sin φ φ/ x + cos φ φ/ y The GST can be constructed from irregular and uncertain data by first stamping the data with certainty onto the SR grid as described in section 3. The image gradients ( I) can then be computed using Normalized Differential Convolution (NDC) [10] or Derivative of Normalized Convolution (DoNC) [17]. Both techniques are capable of estimating gradient from uncertain data. Although NDC is an optimal estimator in a least-squares sense, DoNC is used in this paper due to its simple implementation and a satisfactory result. The gradient scale used in equation 5 for Gaussian smoothing (g) and derivative (g x ) is 1. The tensor smoothing scale ( ) used in equation 3 to compute the GST is 3. DoNC = x ( ) (s.c) g c g = (c g)((s.c) g x) (c g x )((s.c) g) (c g) Structure Adaptive Filtering (5) Local structure adaptive filtering is not a new concept. Several authors [18] [13] have proposed to shape the smoothing kernel after the principal components of the inverse GST. In this paper, we use the same adaptive filtering scheme as in [13]. The filter is a curvature-bent anisotropic Gaussian with its two orthogonal axes aligned with the directions of the GST s principal components. The scales along these directions are given by equation 6, where C encompasses the local signal-to-noise ratio, α controls the degree of structure enhancement and σ dens is the density transform as computed in section 4. In this paper, we use the default value of 1 for both C and α. It is then trivial to see that σ u /σ v = λ v /λ u as suggested in [18]. The novel element of this technique is the use of curvature to enable a better fit of the filter to the local linear structure. σ u = C(1 A) α σ dens σ v = C(1 + A) α σ dens (6) Although the adaptive filter varies at each pixel position, it needs not be inefficient if a right implementation is used. Rather than rotating, scaling and bending the filter to fit with local image, we do the inverse transformation to the local image instead. The anisotropic Gaussian filter could then be looked up from a coefficient table. Furthermore, if the image contains a lot of linear structures, the filter is contracted in one direction therefore few operations are needed for the direct convolution. For example, our implementation 1 of the local structure adaptive filter would take 0.6, 3 and 13 seconds for a 256x256, 512x512 and 1024x1024 image respectively, including time for computing the adaptive parameters. 6 Super-resolution Experiments Every super-resolution algorithm consists of at least two components: image registration and image fusion. With a known registration, the problem reduces to fusion of irregular and uncertain data. The signal irregularity comes from the fact that the scene is captured at different positions in space and time. In fact, the more irregular the data is, the better superresolution result could be obtained. The signal uncertainty comes from the noise of the signal as well as the registration. Registration using optical flow, for example, generates high-confident flow for complex textured regions and low-confident flow for flat regions. Up until now, there are very few superresolution algorithms that incorporate signal certainty into their model. As a result, a single poorly registered frame could totally degrade the output quality. With normalized averaging, however, such disaster could be avoided since we can put less emphasis on these culpable frames. To demonstrate the effectiveness of adaptive normalized averaging for super-resolution, we apply it to two experiments: one with a synthetic sequence and known registration and one with a real sequence taken by a digital camcorder. 6.1 A Synthetic Example In the first experiment, 256x256 super-resolution images are constructed from multiple 64x64 lowresolution images. The LR images are generated from a HR Pentagon scene (1024x1024) using a square sensor model found in common CCD cameras. Difference in image registration is achieved by rotating and translating the scene before a simulated sampling process. Gaussian noise is added to both the gray values and the registered positions. In this experiment, although the registration is known, very few LR frames are available for a 4-time upsampling in 1 C implementation with Matlab interface on an AMD Athlon XP 1.47MHz 1GB RAM
5 each dimension. Besides the noise, a high fill-factor is also an unnegligible factor in all SR construction algorithms (perfect point resampling with 0% fill-factor is often assumed in many experiments in the literature [6] [5]). Despite many of these poor conditions, the performance of SR using adaptive normalized averaging is superior than surface interpolation using Delaunay triangulation [6]. In figure 5a, under a noiseless condition with a small fill-factor, the two methods perform equally well. As soon as noise is present in both gray values as well as registration (figure 5b), adaptive normalized averaging start to outperforms the other method. The effect of noise is largely diminished thanks to an adaptive regularization along local structures. Furthermore, the oriented regularization also requires fewer LR frames for a visually convincing reconstruction because extra local information such as orientation and curvature are used in the fusion process. These adaptive parameters for the Pentagon experiment with 25% fill-factor under a moderate noisy condition can also be seen in figure 4. To further understand why adaptive normalized averaging is less susceptible to noise than a perfect surface reconstruction method, we need to take a closer look at the input and the adaptive parameters. The Pentagon image is a good image for adaptive fusion since the building contains a lot of linear structures. This fact can be confirmed in the orientation image (fig. 4c), in which there are areas of similar color along each of the five facets of the building. As long as the dominant orientation is known, a 2-D superresolution problem reduces to a single-dimensional reconstruction problem. Four LR frames therefore is enough for a 4-by-4 upsampling as shown in figure 5b. The other two parameters to be looked at are the directional scales σ v and σ u. Again, the trace of the Pentagon s five facets is seen in σ v, where a darker value means less smoothing across the building edges. The same region in σ u image gets brighter, on the other hand, which corresponds to an elongated smoothing to suppress noises along the linear structures. Not only adaptive on the orietation, our SR algorithm also adjusts it fusion process based on sample density. The interwoven pattern on both scale images (fig. 4d-e) is the effect of an automatic scale selection based on local density of LR samples. Black dots in between the pattern suggest a presence of LR samples. These positions get less diffusion compared to other locations with larger gaps in the data. As more LR frames are available with correct registration, the SR certainty image may get filled more evenly and this interwowen pattern may be less noticeable. Finally, the result of SR fusion under an extreme situation such as high-interference radiowave imaging is shown in figure 5c. Ten LRs of 64x64 are synthetically captured by a square CCD array with 100% fill-factor. High Gaussian noise are added to both intensity (σ I = 20) and registration of LR images (σ r eg = 0.5 LR pitch). The LR images (fig. 4c) are almost unrecognizable, so does the SR construction using Delaunay triangulation. However, with structure adaptive fusion, valuable structures, such as the path leading into the building, has been successfully recovered. 7 Conclusion In conclusion, we have shown that adaptive normalized averaging is an effective technique for superresolution fusion of image sequences with known registration. The low resolution input samples at irregular positions are incorporated into the super-resolution grid using 4-nearest neighbor stamping with divided sample certainties. Fusion takes place on the superresolution grid by normalized averaging with a structure adaptive applicability. The applicability function is a scale-adaptive, curvature-bent, anisotropic Gaussian that aligns to the local linear structures. The scale of the applicability is derived from the sample density, which minimizes unnecessary blurring in the final fusion. A current limitation of the algorithm is a lack of a built-in deconvolution routine. Our research in the future therefore would be how deblurring could increase the fusion quality, and whether it should be done inline or as a post-processing stage. A modification of the filter shape in the normalized averaging equation is required for such an integration. Under this scheme, it would then be possible to deconvolve spatially variant blur [19], which is commonly found in scenes with a large depth of field. References [1] P. Cheeseman, B. Kanefsky, R. Kraft, and J. Stutz. Super-resolved surface reconstruction from multiple images. Technical report FIA-94-12, NASA, [2] S. Borman and R.L. Stevenson. Superresolution from image sequences - a review. In Proc. of the 1998 Midwest Symposium on circuits and Systems, volume 5, [3] M. Irani and S. Peleg. Improving resolution by image registration. CVGIP: Graphical Models and Image Processing, 53(3): , [4] R.C. Hardie, K.J. Barnard, and E.E. Armstrong. Joint map registration and high-resolution image estimation using a sequence of undersampled images. IEEE Trans. on Image Processing, 6(12): , 1997.
6 a. 64x64 noisy LR input with 25% fill-factor b. very noisy LR input with 100% fill-factor c. Orientation on SR grid (φ [ π/2 π/2]) d. Anisotropy on SR grid (A [0 1]) e. Smoothing scale orientation (σ u ) f. Smoothing scale orientation (σ v ) Figure 4: Control parameters for a 4-time SR from 4 noisy 64x64 LR frames in figure a (images are stretched).
7 a. 4x SR from 20 frames (fill-factor f f = 6.25%, intensity noise σ I = 0, registration noise σ reg = 0) b. 4x SR from 4 noisy frames in figure 4a (f f = 25%, σ I = 10, σ reg = 0.2) c. 4x SR from 10 very noisy frames in figure 4b (f f = 100%, σ I = 20, σ reg = 0.5) Figure 5: 4-time super-resolution in each direction from few noisy 64x64 LR frames with noisy registration (σ reg is measured in LR pitch) using Delaunay triangulation (left) and adaptive normalized averaging (right).
8 [5] S.P. Kim, N.K. Bose, and H.M. Valenzuela. Recursive reconstruction of high resolution image from noisy undersampled multiframes. IEEE Trans. on Acoustics, Speech and Signal Processing, 38: , [6] S. Lertrattanapanich and N.K. Bose. High resolution image formation from low resolution frames using Delaunay triangulation. IEEE TIP, 11(12): , [7] R.Y. Tsai and T.S. Huang. Multiframe Image Resoration and Registration, chapter 7 in: Advances in Computer Vision and Image Processing, pages [8] H. Ur and D. Gross. Improved resolution from subpixel shifted pictures. CVGIP: Graphical Models and Image Processing, 54(2): , [9] A. Okabe, B. Boots, and K. Sugihara. Spatial tessellations Concepts and applications of Voronoi diagrams. Wiley, [10] H. Knutsson and C-F. Westin. Normalized and differential convolution: Methods for interpolation and filtering of incomplete and uncertain data. In Proceedings of IEEE CVPR, pages , New York City, USA, [11] I.T. Young and L.J van Vliet. Recursive implementation of the gaussian filter. Signal Processing, 44(2): , [12] K. Andersson and H. Knutsson. Continuous normalized convolution. In Proceedings of ICME 02, pages , Lausanne, Switzerland, [13] T.Q. Pham and L.J. van Vliet. Normalized averaging using adaptive applicability functions with application in image reconstruction from sparsely and randomly sampled data. In Proceedings of SCIA 03, LNCS Vol. 2749, pages , Goteborg, Sweden, Springer- Verlag. [14] P.E. Danielsson. Euclidean distance mapping. CGIP, 14: , [15] G.H. Granlund. In search for general picture processing operator. CGIP, 8(2): , [16] M. van Ginkel, J. van de Weijer, L.J. van Vliet, and P.W. Verbeek. Curvature estimation from orientation fields. In SCIA 99, pages , Greenland, [17] F. de Jong, L.J. van Vliet, and P.P. Jonker. Gradient estimation in uncertain data. In MVA 98, pages , Makuhari, Chiba, Japan, [18] M. Nitzberg and T. Shiota. Nonlinear image filtering with edge and corner enhancement. IEEE Trans. on PAMI, 14(8): , [19] M-C. Chiang and T.E. Boult. Local blur estimation and super-resolution. In CVPR 97, pages , Puerto Rico, 1997.
Normalized averaging using adaptive applicability functions with applications in image reconstruction from sparsely and randomly sampled data
Normalized averaging using adaptive applicability functions with applications in image reconstruction from sparsely and randomly sampled data Tuan Q. Pham, Lucas J. van Vliet Pattern Recognition Group,
More informationSuper-Resolution on Moving Objects and Background
Super-Resolution on Moving Objects and Background A. van Eekeren K. Schutte J. Dijk D.J.J. de Lange L.J. van Vliet TNO Defence, Security and Safety, P.O. Box 96864, 2509 JG, The Hague, The Netherlands
More informationPerformance study on point target detection using super-resolution reconstruction
Performance study on point target detection using super-resolution reconstruction Judith Dijk a,adamw.m.vaneekeren ab, Klamer Schutte a Dirk-Jan J. de Lange a, Lucas J. van Vliet b a Electro Optics Group
More informationSuper-resolution Image Reconstuction Performance
Super-resolution Image Reconstuction Performance Sina Jahanbin, Richard Naething March 30, 2005 Abstract As applications involving the capture of digital images become more ubiquitous and at the same time
More informationRegion Weighted Satellite Super-resolution Technology
Region Weighted Satellite Super-resolution Technology Pao-Chi Chang and Tzong-Lin Wu Department of Communication Engineering, National Central University Abstract Super-resolution techniques that process
More informationCoarse-to-fine image registration
Today we will look at a few important topics in scale space in computer vision, in particular, coarseto-fine approaches, and the SIFT feature descriptor. I will present only the main ideas here to give
More informationRobust Fusion of Irregularly Sampled Data Using Adaptive Normalized Convolution
Hindawi Publishing Corporation EURASIP Journal on Applied Signal Processing Volume 26, Article ID 8268, Pages 1 12 DOI 1.1155/ASP/26/8268 Robust Fusion of Irregularly Sampled Data Using Adaptive Normalized
More informationMulti-frame super-resolution with no explicit motion estimation
Multi-frame super-resolution with no explicit motion estimation Mehran Ebrahimi and Edward R. Vrscay Department of Applied Mathematics Faculty of Mathematics, University of Waterloo Waterloo, Ontario,
More informationSuper Resolution Using Graph-cut
Super Resolution Using Graph-cut Uma Mudenagudi, Ram Singla, Prem Kalra, and Subhashis Banerjee Department of Computer Science and Engineering Indian Institute of Technology Delhi Hauz Khas, New Delhi,
More informationLocally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling
Locally Weighted Least Squares Regression for Image Denoising, Reconstruction and Up-sampling Moritz Baecher May 15, 29 1 Introduction Edge-preserving smoothing and super-resolution are classic and important
More informationStructured Light II. Thanks to Ronen Gvili, Szymon Rusinkiewicz and Maks Ovsjanikov
Structured Light II Johannes Köhler Johannes.koehler@dfki.de Thanks to Ronen Gvili, Szymon Rusinkiewicz and Maks Ovsjanikov Introduction Previous lecture: Structured Light I Active Scanning Camera/emitter
More informationComputer Vision I - Filtering and Feature detection
Computer Vision I - Filtering and Feature detection Carsten Rother 30/10/2015 Computer Vision I: Basics of Image Processing Roadmap: Basics of Digital Image Processing Computer Vision I: Basics of Image
More informationInvestigation of Superresolution using Phase based Image Matching with Function Fitting
Research Journal of Engineering Sciences ISSN 2278 9472 Investigation of Superresolution using Phase based Image Matching with Function Fitting Abstract Budi Setiyono 1, Mochamad Hariadi 2 and Mauridhi
More informationEE795: Computer Vision and Intelligent Systems
EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 FDH 204 Lecture 14 130307 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Stereo Dense Motion Estimation Translational
More informationNovel Iterative Back Projection Approach
IOSR Journal of Computer Engineering (IOSR-JCE) e-issn: 2278-0661, p- ISSN: 2278-8727Volume 11, Issue 1 (May. - Jun. 2013), PP 65-69 Novel Iterative Back Projection Approach Patel Shreyas A. Master in
More informationCapturing, Modeling, Rendering 3D Structures
Computer Vision Approach Capturing, Modeling, Rendering 3D Structures Calculate pixel correspondences and extract geometry Not robust Difficult to acquire illumination effects, e.g. specular highlights
More information3D Computer Vision. Structured Light II. Prof. Didier Stricker. Kaiserlautern University.
3D Computer Vision Structured Light II Prof. Didier Stricker Kaiserlautern University http://ags.cs.uni-kl.de/ DFKI Deutsches Forschungszentrum für Künstliche Intelligenz http://av.dfki.de 1 Introduction
More informationMulti-dimensional Image Analysis
Multi-dimensional Image Analysis Lucas J. van Vliet www.ph.tn.tudelft.nl/~lucas Image Analysis Paradigm scene Image formation sensor pre-processing Image enhancement Image restoration Texture filtering
More informationMotion Estimation. There are three main types (or applications) of motion estimation:
Members: D91922016 朱威達 R93922010 林聖凱 R93922044 謝俊瑋 Motion Estimation There are three main types (or applications) of motion estimation: Parametric motion (image alignment) The main idea of parametric motion
More informationIntroduction to Image Super-resolution. Presenter: Kevin Su
Introduction to Image Super-resolution Presenter: Kevin Su References 1. S.C. Park, M.K. Park, and M.G. KANG, Super-Resolution Image Reconstruction: A Technical Overview, IEEE Signal Processing Magazine,
More informationcse 252c Fall 2004 Project Report: A Model of Perpendicular Texture for Determining Surface Geometry
cse 252c Fall 2004 Project Report: A Model of Perpendicular Texture for Determining Surface Geometry Steven Scher December 2, 2004 Steven Scher SteveScher@alumni.princeton.edu Abstract Three-dimensional
More informationGRID WARPING IN TOTAL VARIATION IMAGE ENHANCEMENT METHODS. Andrey Nasonov, and Andrey Krylov
GRID WARPING IN TOTAL VARIATION IMAGE ENHANCEMENT METHODS Andrey Nasonov, and Andrey Krylov Lomonosov Moscow State University, Moscow, Department of Computational Mathematics and Cybernetics, e-mail: nasonov@cs.msu.ru,
More informationPeripheral drift illusion
Peripheral drift illusion Does it work on other animals? Computer Vision Motion and Optical Flow Many slides adapted from J. Hays, S. Seitz, R. Szeliski, M. Pollefeys, K. Grauman and others Video A video
More informationSUMMARY: 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 informationINTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET)
INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET) ISSN 0976 6464(Print) ISSN 0976 6472(Online) Volume 3, Issue 3, October- December (2012), pp. 153-161 IAEME: www.iaeme.com/ijecet.asp
More informationBiometrics Technology: Image Processing & Pattern Recognition (by Dr. Dickson Tong)
Biometrics Technology: Image Processing & Pattern Recognition (by Dr. Dickson Tong) References: [1] http://homepages.inf.ed.ac.uk/rbf/hipr2/index.htm [2] http://www.cs.wisc.edu/~dyer/cs540/notes/vision.html
More informationA 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 informationFeature Tracking and Optical Flow
Feature Tracking and Optical Flow Prof. D. Stricker Doz. G. Bleser Many slides adapted from James Hays, Derek Hoeim, Lana Lazebnik, Silvio Saverse, who 1 in turn adapted slides from Steve Seitz, Rick Szeliski,
More informationDownloaded 09/01/14 to Redistribution subject to SEG license or copyright; see Terms of Use at
Random Noise Suppression Using Normalized Convolution Filter Fangyu Li*, Bo Zhang, Kurt J. Marfurt, The University of Oklahoma; Isaac Hall, Star Geophysics Inc.; Summary Random noise in seismic data hampers
More informationEdge and local feature detection - 2. Importance of edge detection in computer vision
Edge and local feature detection Gradient based edge detection Edge detection by function fitting Second derivative edge detectors Edge linking and the construction of the chain graph Edge and local feature
More informationIMAGE RECONSTRUCTION WITH SUPER RESOLUTION
INTERNATIONAL JOURNAL OF RESEARCH IN COMPUTER APPLICATIONS AND ROBOTICS ISSN 2320-7345 IMAGE RECONSTRUCTION WITH SUPER RESOLUTION B.Vijitha 1, K.SrilathaReddy 2 1 Asst. Professor, Department of Computer
More informationA NEW FEATURE BASED IMAGE REGISTRATION ALGORITHM INTRODUCTION
A NEW FEATURE BASED IMAGE REGISTRATION ALGORITHM Karthik Krish Stuart Heinrich Wesley E. Snyder Halil Cakir Siamak Khorram North Carolina State University Raleigh, 27695 kkrish@ncsu.edu sbheinri@ncsu.edu
More informationLecture 4: Spatial Domain Transformations
# Lecture 4: Spatial Domain Transformations Saad J Bedros sbedros@umn.edu Reminder 2 nd Quiz on the manipulator Part is this Fri, April 7 205, :5 AM to :0 PM Open Book, Open Notes, Focus on the material
More informationImproving Spatial Resolution in Exchange of Temporal Resolution in Aliased Image Sequences
Improving Spatial Resolution in Exchange of Temporal Resolution in Aliased Image Sequences Lucas J. van Vliet and Cris L. Luengo Hendriks Pattern Recognition Group of the Department of Applied Physics
More informationSuper resolution: an overview
Super resolution: an overview C Papathanassiou and M Petrou School of Electronics and Physical Sciences, University of Surrey, Guildford, GU2 7XH, United Kingdom email: c.papathanassiou@surrey.ac.uk Abstract
More informationSuper-Resolution. Many slides from Miki Elad Technion Yosi Rubner RTC and more
Super-Resolution Many slides from Mii Elad Technion Yosi Rubner RTC and more 1 Example - Video 53 images, ratio 1:4 2 Example Surveillance 40 images ratio 1:4 3 Example Enhance Mosaics 4 5 Super-Resolution
More informationImage features. Image Features
Image features Image features, such as edges and interest points, provide rich information on the image content. They correspond to local regions in the image and are fundamental in many applications in
More information3D-Orientation Space; Filters and Sampling
3D-Orientation Space; Filters and Sampling Frank G.A. Faas and Lucas J. van Vliet Pattern Recognition Group, Delft University of Technology, Lorentzweg 1, 2628CJ, Delft, The Netherlands, {faas,lucas}@ph.tn.tudelft.nl
More informationPredictive Interpolation for Registration
Predictive Interpolation for Registration D.G. Bailey Institute of Information Sciences and Technology, Massey University, Private bag 11222, Palmerston North D.G.Bailey@massey.ac.nz Abstract Predictive
More informationExperiments with Edge Detection using One-dimensional Surface Fitting
Experiments with Edge Detection using One-dimensional Surface Fitting Gabor Terei, Jorge Luis Nunes e Silva Brito The Ohio State University, Department of Geodetic Science and Surveying 1958 Neil Avenue,
More informationComputer Vision I - Basics of Image Processing Part 1
Computer Vision I - Basics of Image Processing Part 1 Carsten Rother 28/10/2014 Computer Vision I: Basics of Image Processing Link to lectures Computer Vision I: Basics of Image Processing 28/10/2014 2
More informationImage Processing Lecture 10
Image Restoration Image restoration attempts to reconstruct or recover an image that has been degraded by a degradation phenomenon. Thus, restoration techniques are oriented toward modeling the degradation
More informationFeature Detection and Matching
and Matching CS4243 Computer Vision and Pattern Recognition Leow Wee Kheng Department of Computer Science School of Computing National University of Singapore Leow Wee Kheng (CS4243) Camera Models 1 /
More informationFiltering Images. Contents
Image Processing and Data Visualization with MATLAB Filtering Images Hansrudi Noser June 8-9, 010 UZH, Multimedia and Robotics Summer School Noise Smoothing Filters Sigmoid Filters Gradient Filters Contents
More informationSchedule 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 informationFPGA-based Real-time Super-Resolution on an Adaptive Image Sensor
FPGA-based Real-time Super-Resolution on an Adaptive Image Sensor Maria E. Angelopoulou, Christos-Savvas Bouganis, Peter Y. K. Cheung, and George A. Constantinides Department of Electrical and Electronic
More informationEnhancing DubaiSat-1 Satellite Imagery Using a Single Image Super-Resolution
Enhancing DubaiSat-1 Satellite Imagery Using a Single Image Super-Resolution Saeed AL-Mansoori 1 and Alavi Kunhu 2 1 Associate Image Processing Engineer, SIPAD Image Enhancement Section Emirates Institution
More informationRobotics 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 informationApplication of Tatian s Method to Slanted-Edge MTF Measurement
Application of s Method to Slanted-Edge MTF Measurement Peter D. Burns Eastman Kodak Company, Rochester, NY USA 465-95 ABSTRACT The 33 method for the measurement of the spatial frequency response () of
More informationColor Image Segmentation
Color Image Segmentation Yining Deng, B. S. Manjunath and Hyundoo Shin* Department of Electrical and Computer Engineering University of California, Santa Barbara, CA 93106-9560 *Samsung Electronics Inc.
More informationLaser sensors. Transmitter. Receiver. Basilio Bona ROBOTICA 03CFIOR
Mobile & Service Robotics Sensors for Robotics 3 Laser sensors Rays are transmitted and received coaxially The target is illuminated by collimated rays The receiver measures the time of flight (back and
More informationScanner Parameter Estimation Using Bilevel Scans of Star Charts
ICDAR, Seattle WA September Scanner Parameter Estimation Using Bilevel Scans of Star Charts Elisa H. Barney Smith Electrical and Computer Engineering Department Boise State University, Boise, Idaho 8375
More informationComputer Vision 2. SS 18 Dr. Benjamin Guthier Professur für Bildverarbeitung. Computer Vision 2 Dr. Benjamin Guthier
Computer Vision 2 SS 18 Dr. Benjamin Guthier Professur für Bildverarbeitung Computer Vision 2 Dr. Benjamin Guthier 3. HIGH DYNAMIC RANGE Computer Vision 2 Dr. Benjamin Guthier Pixel Value Content of this
More informationconvolution shift invariant linear system Fourier Transform Aliasing and sampling scale representation edge detection corner detection
COS 429: COMPUTER VISON Linear Filters and Edge Detection convolution shift invariant linear system Fourier Transform Aliasing and sampling scale representation edge detection corner detection Reading:
More informationx' = c 1 x + c 2 y + c 3 xy + c 4 y' = c 5 x + c 6 y + c 7 xy + c 8
1. Explain about gray level interpolation. The distortion correction equations yield non integer values for x' and y'. Because the distorted image g is digital, its pixel values are defined only at integer
More informationCS 565 Computer Vision. Nazar Khan PUCIT Lectures 15 and 16: Optic Flow
CS 565 Computer Vision Nazar Khan PUCIT Lectures 15 and 16: Optic Flow Introduction Basic Problem given: image sequence f(x, y, z), where (x, y) specifies the location and z denotes time wanted: displacement
More informationIMPROVED MOTION-BASED LOCALIZED SUPER RESOLUTION TECHNIQUE USING DISCRETE WAVELET TRANSFORM FOR LOW RESOLUTION VIDEO ENHANCEMENT
17th European Signal Processing Conference (EUSIPCO 009) Glasgow, Scotland, August 4-8, 009 IMPROVED MOTION-BASED LOCALIZED SUPER RESOLUTION TECHNIQUE USING DISCRETE WAVELET TRANSFORM FOR LOW RESOLUTION
More informationComputer Vision I - Basics of Image Processing Part 2
Computer Vision I - Basics of Image Processing Part 2 Carsten Rother 07/11/2014 Computer Vision I: Basics of Image Processing Roadmap: Basics of Digital Image Processing Computer Vision I: Basics of Image
More informationA Survey On Super Resolution Image Reconstruction Techniques
A Survey On Super Resolution Image Reconstruction Techniques Krunal Shah Jaymit Pandya Safvan Vahora Dept. of Info Tech,GCET Dept. of Info Tech,GCET Dept. of Info Tech,VGEC Engg.College, VV Nagar, Engg,College,VV
More informationFourier Transform and Texture Filtering
Fourier Transform and Texture Filtering Lucas J. van Vliet www.ph.tn.tudelft.nl/~lucas Image Analysis Paradigm scene Image formation sensor pre-processing Image enhancement Image restoration Texture filtering
More informationSuper-Resolution from Image Sequences A Review
Super-Resolution from Image Sequences A Review Sean Borman, Robert L. Stevenson Department of Electrical Engineering University of Notre Dame 1 Introduction Seminal work by Tsai and Huang 1984 More information
More informationGeneration of Triangle Meshes from Time-of-Flight Data for Surface Registration
Generation of Triangle Meshes from Time-of-Flight Data for Surface Registration Thomas Kilgus, Thiago R. dos Santos, Alexander Seitel, Kwong Yung, Alfred M. Franz, Anja Groch, Ivo Wolf, Hans-Peter Meinzer,
More informationA Frequency Domain Approach to Super-Resolution Imaging from Aliased Low Resolution Images
A Frequency Domain Approach to Super-Resolution Imaging from Aliased Low Resolution Images Patrick Vandewalle, Student Member, IEEE, Sabine Süsstrunk, Member, IEEE, and Martin Vetterli, Fellow, IEEE Abstract
More informationAdaptive Kernel Regression for Image Processing and Reconstruction
Adaptive Kernel Regression for Image Processing and Reconstruction Peyman Milanfar* EE Department University of California, Santa Cruz *Joint work with Sina Farsiu, Hiro Takeda AFOSR Sensing Program Review,
More informationImage Processing Fundamentals. Nicolas Vazquez Principal Software Engineer National Instruments
Image Processing Fundamentals Nicolas Vazquez Principal Software Engineer National Instruments Agenda Objectives and Motivations Enhancing Images Checking for Presence Locating Parts Measuring Features
More informationTexture. 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 informationCourse Evaluations. h"p:// 4 Random Individuals will win an ATI Radeon tm HD2900XT
Course Evaluations h"p://www.siggraph.org/courses_evalua4on 4 Random Individuals will win an ATI Radeon tm HD2900XT A Gentle Introduction to Bilateral Filtering and its Applications From Gaussian blur
More informationRobust and Accurate Detection of Object Orientation and ID without Color Segmentation
0 Robust and Accurate Detection of Object Orientation and ID without Color Segmentation Hironobu Fujiyoshi, Tomoyuki Nagahashi and Shoichi Shimizu Chubu University Japan Open Access Database www.i-techonline.com
More informationFace Hallucination Based on Eigentransformation Learning
Advanced Science and Technology etters, pp.32-37 http://dx.doi.org/10.14257/astl.2016. Face allucination Based on Eigentransformation earning Guohua Zou School of software, East China University of Technology,
More informationSEMI-BLIND IMAGE RESTORATION USING A LOCAL NEURAL APPROACH
SEMI-BLIND IMAGE RESTORATION USING A LOCAL NEURAL APPROACH Ignazio Gallo, Elisabetta Binaghi and Mario Raspanti Universitá degli Studi dell Insubria Varese, Italy email: ignazio.gallo@uninsubria.it ABSTRACT
More informationLocal Image Features
Local Image Features Computer Vision CS 143, Brown Read Szeliski 4.1 James Hays Acknowledgment: Many slides from Derek Hoiem and Grauman&Leibe 2008 AAAI Tutorial This section: correspondence and alignment
More informationIMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS
IMAGE DENOISING TO ESTIMATE THE GRADIENT HISTOGRAM PRESERVATION USING VARIOUS ALGORITHMS P.Mahalakshmi 1, J.Muthulakshmi 2, S.Kannadhasan 3 1,2 U.G Student, 3 Assistant Professor, Department of Electronics
More informationFeature Based Registration - Image Alignment
Feature Based Registration - Image Alignment Image Registration Image registration is the process of estimating an optimal transformation between two or more images. Many slides from Alexei Efros http://graphics.cs.cmu.edu/courses/15-463/2007_fall/463.html
More informationMotion Estimation and Optical Flow Tracking
Image Matching Image Retrieval Object Recognition Motion Estimation and Optical Flow Tracking Example: Mosiacing (Panorama) M. Brown and D. G. Lowe. Recognising Panoramas. ICCV 2003 Example 3D Reconstruction
More informationSegmentation and Grouping
Segmentation and Grouping How and what do we see? Fundamental Problems ' Focus of attention, or grouping ' What subsets of pixels do we consider as possible objects? ' All connected subsets? ' Representation
More informationJunction Detection and Multi-orientation Analysis Using Streamlines
Junction Detection and Multi-orientation Analysis Using Streamlines Frank G.A. Faas and Lucas J. van Vliet Quantitative Imaging Group, Delft University of Technology, The Netherlands L.J.vanVliet@tudelft.nl
More informationImproving Latent Fingerprint Matching Performance by Orientation Field Estimation using Localized Dictionaries
Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC, Vol. 3, Issue. 11, November 2014,
More informationMulti-stable Perception. Necker Cube
Multi-stable Perception Necker Cube Spinning dancer illusion, Nobuyuki Kayahara Multiple view geometry Stereo vision Epipolar geometry Lowe Hartley and Zisserman Depth map extraction Essential matrix
More informationLecture 7: Most Common Edge Detectors
#1 Lecture 7: Most Common Edge Detectors Saad Bedros sbedros@umn.edu Edge Detection Goal: Identify sudden changes (discontinuities) in an image Intuitively, most semantic and shape information from the
More informationSECTION 5 IMAGE PROCESSING 2
SECTION 5 IMAGE PROCESSING 2 5.1 Resampling 3 5.1.1 Image Interpolation Comparison 3 5.2 Convolution 3 5.3 Smoothing Filters 3 5.3.1 Mean Filter 3 5.3.2 Median Filter 4 5.3.3 Pseudomedian Filter 6 5.3.4
More informationLevel lines based disocclusion
Level lines based disocclusion Simon Masnou Jean-Michel Morel CEREMADE CMLA Université Paris-IX Dauphine Ecole Normale Supérieure de Cachan 75775 Paris Cedex 16, France 94235 Cachan Cedex, France Abstract
More informationA Keypoint Descriptor Inspired by Retinal Computation
A Keypoint Descriptor Inspired by Retinal Computation Bongsoo Suh, Sungjoon Choi, Han Lee Stanford University {bssuh,sungjoonchoi,hanlee}@stanford.edu Abstract. The main goal of our project is to implement
More informationMotion and Optical Flow. Slides from Ce Liu, Steve Seitz, Larry Zitnick, Ali Farhadi
Motion and Optical Flow Slides from Ce Liu, Steve Seitz, Larry Zitnick, Ali Farhadi We live in a moving world Perceiving, understanding and predicting motion is an important part of our daily lives Motion
More informationInternational ejournals
ISSN 2249 5460 Available online at www.internationalejournals.com International ejournals International Journal of Mathematical Sciences, Technology and Humanities 96 (2013) 1063 1069 Image Interpolation
More informationComparative Analysis of Edge Based Single Image Superresolution
Comparative Analysis of Edge Based Single Image Superresolution Sonali Shejwal 1, Prof. A. M. Deshpande 2 1,2 Department of E&Tc, TSSM s BSCOER, Narhe, University of Pune, India. ABSTRACT: Super-resolution
More informationFactorization with Missing and Noisy Data
Factorization with Missing and Noisy Data Carme Julià, Angel Sappa, Felipe Lumbreras, Joan Serrat, and Antonio López Computer Vision Center and Computer Science Department, Universitat Autònoma de Barcelona,
More informationLocally Adaptive Regression Kernels with (many) Applications
Locally Adaptive Regression Kernels with (many) Applications Peyman Milanfar EE Department University of California, Santa Cruz Joint work with Hiro Takeda, Hae Jong Seo, Xiang Zhu Outline Introduction/Motivation
More informationIMAGE DE-NOISING IN WAVELET DOMAIN
IMAGE DE-NOISING IN WAVELET DOMAIN Aaditya Verma a, Shrey Agarwal a a Department of Civil Engineering, Indian Institute of Technology, Kanpur, India - (aaditya, ashrey)@iitk.ac.in KEY WORDS: Wavelets,
More informationAn adaptive preprocessing algorithm for low bitrate video coding *
Li et al. / J Zhejiang Univ SCIENCE A 2006 7(12):2057-2062 2057 Journal of Zhejiang University SCIENCE A ISSN 1009-3095 (Print); ISSN 1862-1775 (Online) www.zju.edu.cn/jzus; www.springerlink.com E-mail:
More informationSampling and Reconstruction
Sampling and Reconstruction Sampling and Reconstruction Sampling and Spatial Resolution Spatial Aliasing Problem: Spatial aliasing is insufficient sampling of data along the space axis, which occurs because
More informationEFFICIENT PERCEPTUAL, SELECTIVE,
EFFICIENT PERCEPTUAL, SELECTIVE, AND ATTENTIVE SUPER-RESOLUTION RESOLUTION Image, Video & Usability (IVU) Lab School of Electrical, Computer, & Energy Engineering Arizona State University karam@asu.edu
More informationMatching. Compare region of image to region of image. Today, simplest kind of matching. Intensities similar.
Matching Compare region of image to region of image. We talked about this for stereo. Important for motion. Epipolar constraint unknown. But motion small. Recognition Find object in image. Recognize object.
More informationMulti-Orientation Estimation: Selectivity and Localization
in: M. van Ginkel, P.W. Verbeek, and L.J. van Vliet, Multi-orientation estimation: Selectivity and localization, in: H.E. Bal, H. Corporaal, P.P. Jonker, J.F.M. Tonino (eds.), ASCI 97, Proc. 3rd Annual
More informationPoint-Based Rendering
Point-Based Rendering Kobbelt & Botsch, Computers & Graphics 2004 Surface Splatting (EWA: Elliptic Weighted Averaging) Main Idea Signal Processing Basics Resampling Gaussian Filters Reconstruction Kernels
More informationCS5670: Computer Vision
CS5670: Computer Vision Noah Snavely Lecture 4: Harris corner detection Szeliski: 4.1 Reading Announcements Project 1 (Hybrid Images) code due next Wednesday, Feb 14, by 11:59pm Artifacts due Friday, Feb
More informationComputer Vision. Recap: Smoothing with a Gaussian. Recap: Effect of σ on derivatives. Computer Science Tripos Part II. Dr Christopher Town
Recap: Smoothing with a Gaussian Computer Vision Computer Science Tripos Part II Dr Christopher Town Recall: parameter σ is the scale / width / spread of the Gaussian kernel, and controls the amount of
More informationDisparity from Monogenic Phase
Disparity from Monogenic Phase Michael Felsberg Department of Electrical Engineering, Linköping University, SE-58183 Linköping, Sweden mfe@isy.liu.se, WWW home page: http://www.isy.liu.se/~mfe Abstract.
More informationComputer Vision for HCI. Topics of This Lecture
Computer Vision for HCI Interest Points Topics of This Lecture Local Invariant Features Motivation Requirements, Invariances Keypoint Localization Features from Accelerated Segment Test (FAST) Harris Shi-Tomasi
More informationEECS 556 Image Processing W 09. Interpolation. Interpolation techniques B splines
EECS 556 Image Processing W 09 Interpolation Interpolation techniques B splines What is image processing? Image processing is the application of 2D signal processing methods to images Image representation
More informationA Fast Super-Resolution Reconstruction Algorithm for Pure Translational Motion and Common Space-Invariant Blur
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 10, NO. 8, AUGUST 2001 1187 A Fast Super-Resolution Reconstruction Algorithm for Pure Translational Motion and Common Space-Invariant Blur Michael Elad, Member,
More information