A Learning Based Method for Super-Resolution of Low Resolution Images

Size: px
Start display at page:

Download "A Learning Based Method for Super-Resolution of Low Resolution Images"

Transcription

1 A Learning Based Method for Super-Resolution of Low Resolution Images Emre Ugur June 1, 2004 Abstract The main objective of this project is the study of a learning based method for super-resolving low resolution images. The domain specific prior is incorporated into super resolution by the means of learning based estimation of missing details. Images are decomposed into fixed size patches in order to deal with time and space complexity. The problem is modeled by Markov Random Field which enforces resulting images to be spatially consistent. The spatial interactions are coupled with a similarity constraint which should be established between high-resolution training image patches and low resolution observations. 1 Introduction The main objective of this project is the study of a learning based method for super-resolving low resolution images. In super resolution methods, given a low resolution image or image sequence (a video), one tries to estimate the highly zoomed image or video, and extract the missing image details. Super resolution problem is an ill-posed inverse problem. Estimating details is an inverse problem since low resolution observation is the result of a smoothing and downsampling process. Additionally the problem is ill-posed because a large amount of information is lost and there exists a number of high resolution images which result in same low resolution image, when they are smoothed and downsampled. In the past two decades, the problem of estimation of fine details from low resolution images has attracted a number of researchers whose approaches can be classified in two main classes. First and the more common one is Reconstruction Based (RB) methods [5] which endeavor increasing the effective sampling density. These methods attempt to solve the problem by employing and fusing a number of low-resolution images. The images are of an underlying scene are positioned into a common coordinate frame by sub-pixel shifts of images, and a coherent high resolution image is obtained in a finer resolution grid. The approach employed in this report, locates in a relatively recent solution family, which is based on Learning-based techniques. A training set, which is composed of a number of high-resolution images/video, is used to later predict the details of lower resolution images/video. Both classes employ additional priors to 1

2 encourage generic image properties such as local smoothness, implicitly or explicitly. Although in RB, multiple low resolution observations of the same scene is required to estimate the high resolution counterpart, in Learning-based methods, it is proposed that only one low resolution image includes adequate information to predict the details and to super resolve the image. In [1], Super resolution is defined as a process whose goal is to increase the resolution and at the same time adding appropriate high frequency information. Therefore, [1] employs a large database of image pairs, which stores a rectangular patch of high frequency component a high resolution image and corresponding smoothed and downsampled low level counter-part. The relationship between middle and high frequency of natural images are captured and used to super resolve low resolution static images and movies. Although a zoom factor of x4 is established for static images, direct application of the approach is not successful in video. In [2], spatio-temporal consistencies and image formation/degradation processes are employed to estimate or hallucinate high resolution video. Since learning based approaches are more powerful when their application is limited to a specific domain, a database of facial expressions is obtained using a sequence (video) of high resolution images. Their low resolution counterpart is acquired using a local smoothing and downsampling process. Images are divided into patches, and spatial (via single image) and temporal (via time) consistencies are established among these image patches. After the training database is constructed, they tries to find Maximum A Priori high resolution image and illumination offset by finding out a template image. Template image is constructed from high resolution patches in the database, maximizing some constraints. Since this project is based on the approach of [2], the details of model and algorithm will be described in preceding sections. 1.1 Differences from original work The assumptions employed in this study differ from [2] in different aspects. First of all we assume that the illumination conditions are similar or same in high resolution training set data and low resolution observations. Thus, the problem is reduced to only estimate the high resolution image, simplifying the solution. In fact, the hallucination step of [2] is dropped, where a gradient descent algorithm is applied to solve a quadratic minimization problem. This reduction does not affect the importance of our solution since the most important part of the learning-based method, is to find a feasible template image. Other difference appears in registration of training and test data. In [2], the image is adjusted such that the outline of the face is in same coordinates in the database. Therefore, they avoid the unrealistic patch compositions in translational and rotational sense. In our experiments, we do not spend additional effort to adjust face location, and expect the Markovian Model avoid such an effect by forcing similar patches to become neighborhood. The experiments in [2] are mostly based on video, and they well employed the temporal prior or relation between consecutive images. Moreover, they tried to acquire realistic facial expressions from low resolution video where facial expressions change very slowly. In our experiments, our criteria of success is mostly based on the difference from real high resolution image of low resolution observation. In order to register large number of different facial expressions and head 2

3 placements to training set, our consecutive images are used as static images, where temporal relation between images are not established. Thus, experiments are performed only for super resolution of static images. The flow of this report will be as follows. In Section 2, the framework which learning based super resolution algorithm based on is described. A sequence of 75 images, that are captured in 192x304 resolution, are utilized to construct the facial expressions database. A number of artificially downsampled low resolution images are employed to test the system, in order to super resolve with a zoom factor of x16 in Section 3. Additionally, some parameters are tuned in the algorithm and model is modified in order to understand the dynamics of the system in various low resolution test images. In Section 4, conclusion of the study and future work is presented. 2 Framework The main motivation behind this graphical is the complementary usage of low resolution image data via a similarity measure and relations constructed among the training image patches via Markov Random Field. In the very first step, a database is constructed using a number of (in our case a sequence of) high resolution images. Building a database, that will contain all possible facial expressions and pixel combinations, is a very challenging task because of its enormous size and time requirement. Therefore, local models which are defined over image patches and considered independent of each other can be employed in order to define whole image scene. However, unrealistic face compositions may be created if no relation is defined among these local models. Therefore, in order to model spatial relations among patches, Markov Random Field is employed. The details of the statistical relationship established among patches will be described later. 2.1 Registration of training set Training database is composed of entities, which span different properties of the underlying image patch. More specifically, each entry contains (i) corresponding high resolution image patch, (ii) location of the patch in sample image, (iii) neighboring pixels (in order to establish realistic estimations among patches), and (iv) a feature vector [3] which stores different properties of patches low-resolution counterpart. Figure shows the contents of an entry. The size of the high-resolution patch depends on the zoom factor; for a zoom factor of x16, it is defined over a 16x16 grid. The creation and meaning of neighboring pixels is clear, they are used to compute the statistical local relation of neighboring pixels in MRF. The width of neighboring pixels may vary according to the prior of application domain. In order to achieve a more consistent solution in terms of smoothness of neighboring patches, it is desirable to store a wider neighborhood pixel array in entry. Feature vector or parent vector is first defined in [3] in order to define or store different properties of an image region in an hierarchical manner. More specifically, for any pixel, intensity pyramid, Gaussian pyramid, Laplacian pyramid, Horizontal and 1 This image and explanation is taken from [2]. 3

4 Figure 1: Database is composed of entries which are generated from image patches in the training set. Each entry contains the high resolution image patch, neighboring pixels (to be used in MRF interactions), a feature vector of low resolution counterpart, and the location in x-y coordinates. Vertical derivative pyramids are formed in a multi scale. Thus, feature vector is the combination of these pyramids, which provide information about both local features and general underlying image scene. In later steps, feature vector is employed to compute the similarity between different low-resolution image patches from training and test set. 2.2 Estimation process The estimation process is based on finding a unique template from which high resolution image is extracted. In the original work, as mentioned in Section 1.1, high resolution image and an intensity offset were two unknowns of the problem. In this project, we decrease search space, and only search for high resolution image, assuming there is no intensity offset (ie. training and test images are captured in similar conditions). Therefore, super resolution problem maps to finding Maximum A Posteriori (MAP) high resolution image H MAP : H MAP arg max HlogP (H L) where L is the low resolution observation. Above equation is marginalized over unknown template image T, which is composed of image patches in the training database. Therefore, P (H L) = T P (H, T L) If chain rule is applied to the above probabilistic formula, P (H L) = T P (H T, L)P (T L) is obtained. Bayes rules is used to obtain, P (H L) = { } P (L H, T )P (H T ) P (T L) P (L H) T Since we use only Markov Random Field model to relate different nodes, and there is no relation between two nodes which are no linked via MRF, no direct conditioning exists among such nodes. As a result P (L T, L) = P (L H), P (H L) = { } (L H, T )P (H T )P (T L) T 4

5 Since P (T L) has its maximum value around true high resolution solution, we will try to compute a unique (peak) template T to maximize the posterior using low resolution images and database entries. If posterior of P (T L) is approximated to highly concentrate around T = T (L), original posterior which is tried to be maximized, P (H L) becomes, P (H L) = P (L H)P (H T ) Therefore, Maximum a Posteriori high resolution image will be computed by, H MAP arg max{logp (L H)logP (H T )} In Section 2.3, the computation of peak template T will be presented. In the scope of this project, no further estimation step is taken by the means of any hallucination technique. Interested readers can look at Kanade et. al. ([4]) and Dedeoglu et. al. ([2]) for minimization processes employed to satisfy certain criteria. 2.3 Formation of unique template In previous section, it is described that in order to maximize P (T L), a unique template should be constructed from patches in database. Since nodes in L are conditionally dependent, Bayes rule can be applied to in maximization of P (T L), P (T L) P (L T )P (T ) = N P (L p T p)p (T ) (1) The peak template will be computed according to above formulation. By maximizing the first term of right hand side of the equation, the difference between low resolution observation and downsampled unknown template is minimized. Second term of right hand side provides a consistent template, where spatial consistency in the data is established by means of MRF modeling. p= Feature vector and maximizing similarity As defined in Section 2.1, feature vector is the means of storing low resolution patch properties. For training images, high resolution images are smoothed and downsampled to obtain low-resolution feature images. Each entry in the database is a quadruple, (t k, η k, f k, s k ), where, t k is high-resolution template patch, η k is neighboring pixels, f k is the feature vector computed at corresponding low-resolution pixel, and s k is the location of the patch, which is defined in x-y coordinates. In Equation 1, P (F T ) = N p=1 P (P f T p) is maximized by computing the similarity between different feature vectors of template patches and feature vector of low-resolution observation s patch. The location of these patches should be same in order to be comparable. In other words, for each entry k, a similarity measure P (F p = f p, T p = t k ) is computed for each location s k. f p is the feature vector of low resolution observation at location p = s k. P (F p = f p T p = t k ) { exp( fp f k ) if s k = p 0 ow. (2) 5

6 2.3.2 Spatial consistency and relationship among nodes in MRF For each overlapping high resolution patch pair in the training database, a likelihood computation, P (T p, T q), is performed to force the parches spatially consistent. Patched from different images or same image are merged, preserving their location relative to whole scene to form template patch configurations. Each template patch configuration is given a probability which defines the total spatial consistency of corresponding configuration. For example, if template is composed of patches from same image, the probability will be maximum. On the contrary, if neighboring patches of the template are not smooth in their joining points, the probability will be low. The probability of spatial consistency of template T, is the product of each pair inside the configuration, P (T ) = T p,t q (T p, T q). The compatibility function (T p, T q) uses the difference measure between intensity values of overlapping regions, ( (T p, T q) exp overlap (t k(u) n l (v)) 2 overlap (n k(u) t l (v)) 2 ) (3) Computation of T Using equations 2 and 3, T is computed by maximizing, arg max T N P (P f T p) p=1 T p,t q (T p, T q) (4) The computational complexity of searching for optimum solution is huge, so we should compute maximum of the above probabilistic formula by means of an approximate method. Iterated Conditional Modes (ICM) algorithm first constructs a feasible initial template, than performs random walk in search space, maximizing conditional probability defined over MRF. The algorithm is copied from [2], Algorithm 1 ICM Algorithm for T for all image patches p do T p arg max tk P (F p = f p T p = t k ) end for while not converged do T p arg max tk P (F p = f p T p = t k ) q Neighbors(p) (T p = t k, T q) end while 3 Experiments 3.1 Experiment parameters The database of facial expressions is generated from a 150 frame-long video of of a person who continuously and abruptly changes his expressions. As in [2], no ad- 6

7 ditional work is done to adjust the head to a common coordinate frame. However, the person whose video is captured tried not to move his head. The resolution of the training image is 192x304. Although images are taken as video, frames are behaved as static images, and no temporal relation among any image is incorporated. In order to construct the database, high resolution images are smoothed via a local smoothing operator, then a simple downsampling process is employed to obtain the low resolution (12x19) counterparts. The learning-based method, that is employed in this project, is mainly based on two important concepts, maximizing similarity and spatial consistency. Similarity measure is computed by taking difference of feature vectors of different low resolution image patches. Euclidean distance is used to take the distance of feature vectors, each of which is composed of intensity, Laplacian, horizontal and vertical derivatives. The effect of each variable will be examined later in this section. Other important objective is to establish spatial relationship between image patches. As described in Section in detail, Markov Random Field use neighborhood relation to compute the likelihood of any template configuration. Width of the neighborhood pixels is a very important parameter when defining the relative effect of spatial consistency on the solution. In our study, we take the pixel width 1. The real frame length of our video is 250. The last 50 images of the same video are registered as test images. Same video instance is used to extract test images because illumination conditions should be similar and head should be in similar coordinates in the underlying scene. It is almost impossible for any training set and test images to be same, because the subject changes his facial expressions unrealistically, and there is a 50 frame long period between two sets. 3.2 Results As it is demonstrated in Figure 2, learning-based super resolution method gives feasible solutions when a zooming factor of x16 is applied. The images of Figures 3(a) are samples from high resolution test image set. Test images are selected in order to show results of different facial expressions. High resolution images are smoothed and downsampled, in order obtain low resolution observations which serve as inputs to experiments (Figures 3(b)). If only feature vector similarity is employed in order to maximize the likelihood of any template configuration, spatially inconsistent solutions are obtained as shown in Figures 3(c). Since patches in the template behave independently, scene seems to be composed of different image blocks which are not similar in their overlapping areas. Figures 3(d) shows estimated high resolution images which seems better, more fluent since spatial interactions are enforced via Markov Random Field modeling. More specifically, while Figures 3(c) is the result of first step of ICM algorithm (in Section 2.3.3), in Figures 3(d) algorithm is completed. In Figure 3, enlarged version of the first image of Figures 3(c) and 3(d) shows better the effect of incorporating spacial interactions. Since images show relatively smooth patches in experiments, one important question arises, whether template images are composed of patches that belongs to small number of images. Figure 4 demonstrates that, although for each different template, there exists a small number of dominant images in the database, in general most of the images affect the generated template. Figure 5(a) shows the number of patches from each image which which is inserted to the template. In 7

8 Figure 5(b), each patch is painted with a different gray-level if its corresponding image is different than others. This figure clearly shows that different regions (or patches) of the template face is taken from different images in the database. After main results are obtained, next experiments are performed in order to understand the effect of different variables that are stored in feature vector. Remind that our feature vector is composed of intensity, laplacian, horizontal and vertical derivatives. Figure 6(b) shows the resulting template when laplacian component is deleted, Figure 6(c) is the result of omission of derivative components and in Figure 6(a), only intensity component is used. Figures show that horizontal and vertical derivative components are essential in feature vector but there is not much difference between with or without laplacian case. 4 Discussion and Future Work In this project, super resolution problem for a domain specific instance is examined, and feasible results are obtained. The power of learning based methods on domain limited applications is employed in order to achieve a zoom factor of x16. Patches of a training set of static images are registered into a database, storing their local properties and relationships. Markov Random Field is used to model the patch relations, leading to only local interactions. Patches from different images are fused in different combinations, and a probability for each template is computed employing neighborhood relations or overlapping similarities of each patch pair in the template configurations. Therefore, the objective becomes to maximize total of probability of each template configuration and similarity of the template with low resolution observation. Similarity of any template with low resolution observation is computed via Euclidean distance which is defined over feature vectors of the image patches. The results are feasible but far from being perfect. In different figures, one can easily see that image patches do not fit well in their neighborhood relation, even spatial interaction is enforced. There is a trade-off, enforcing spatial interaction and at the same time remaining similar to observation. One can increase pixel width, which is taken 1 in our project, to avoid from patchy images. Another solution may be the usage of varying size patches. Although the time and space complexity increase will be enormous, more fluent images can be obtained, at the same time, preserving similarity to low resolution observation References [1] C. M. Bishop, A. Blake, and B. Marthi, Super-resolution enhancement of video, In C. M. Bishop and B. Frey, editors, Proceedings Artificial Intelligence and Statistics. Society for Artificial Intelligence and Statistics, [2] G. Dedeoglu, T. Kanade, and J. August, High-Zoom Video Hallucination by Exploiting Spatio-Temporal Regularities Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR 04), June, [3] DeBonet and P. A. Viola. A non-parametric multi-scale statistical model for natural images, In Advances in Neural Information Processing Systems (NIPS), volume 10. The MIT Press,

9 [4] S. Baker and T. Kanade. Limits on super-resolution and how to break them, IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(9): , [5] A. Papoulis, Generalized sampling theorem, IEEE Transactions on Circuits and Systems, vol. 24, pp , November

10 Figure 2: (a) High resolution test images in scale 192x304 (b) Low resolution observations downsampled from above in scale 12x19 (c) Template high resolution estimate without spatial interaction 10

11 Figure 3: (a) Enlarged of Figure 3(c) without spacial interaction 11 (b) Enlarged of Figure 3(c) with spacial interaction

12 Figure 4: Although there exists dominant images, templates tend to take patches from almost all images. Number of times images appeared in the template Number of times images appeared in the template Number of times images appeared in the template Number of times images appeared in the template Image Index in Training Set Image Index in Training Set Image Index in Training Set Image Index in Training Set (a) Number of images in the template configuration for Figure 2 (b) Different gray-levels assigned for different image patches for Figure 2 Figure 5: Feature vector is modified (b) No laplacian component. (a) No modification. (c) No vertical or horizontal derivative component. (d) Only intensity component. 12

Bayesian Image Super-Resolution

Bayesian Image Super-Resolution Bayesian Image Super-Resolution Michael E. Tipping and Christopher M. Bishop Microsoft Research, Cambridge, U.K..................................................................... Published as: Bayesian

More information

Extracting a Good Quality Frontal Face Images from Low Resolution Video Sequences

Extracting a Good Quality Frontal Face Images from Low Resolution Video Sequences Extracting a Good Quality Frontal Face Images from Low Resolution Video Sequences Pritam P. Patil 1, Prof. M.V. Phatak 2 1 ME.Comp, 2 Asst.Professor, MIT, Pune Abstract The face is one of the important

More information

High Quality Image Magnification using Cross-Scale Self-Similarity

High Quality Image Magnification using Cross-Scale Self-Similarity High Quality Image Magnification using Cross-Scale Self-Similarity André Gooßen 1, Arne Ehlers 1, Thomas Pralow 2, Rolf-Rainer Grigat 1 1 Vision Systems, Hamburg University of Technology, D-21079 Hamburg

More information

SUPER RESOLUTION FROM MULTIPLE LOW RESOLUTION IMAGES

SUPER RESOLUTION FROM MULTIPLE LOW RESOLUTION IMAGES SUPER RESOLUTION FROM MULTIPLE LOW RESOLUTION IMAGES ABSTRACT Florin Manaila 1 Costin-Anton Boiangiu 2 Ion Bucur 3 Although the technology of optical instruments is constantly advancing, the capture of

More information

Tracking Moving Objects In Video Sequences Yiwei Wang, Robert E. Van Dyck, and John F. Doherty Department of Electrical Engineering The Pennsylvania State University University Park, PA16802 Abstract{Object

More information

Feature Tracking and Optical Flow

Feature Tracking and Optical Flow 02/09/12 Feature Tracking and Optical Flow Computer Vision CS 543 / ECE 549 University of Illinois Derek Hoiem Many slides adapted from Lana Lazebnik, Silvio Saverse, who in turn adapted slides from Steve

More information

Image Super-Resolution for Improved Automatic Target Recognition

Image Super-Resolution for Improved Automatic Target Recognition Image Super-Resolution for Improved Automatic Target Recognition Raymond S. Wagner a and Donald Waagen b and Mary Cassabaum b a Rice University, Houston, TX b Raytheon Missile Systems, Tucson, AZ ABSTRACT

More information

Face Model Fitting on Low Resolution Images

Face Model Fitting on Low Resolution Images Face Model Fitting on Low Resolution Images Xiaoming Liu Peter H. Tu Frederick W. Wheeler Visualization and Computer Vision Lab General Electric Global Research Center Niskayuna, NY, 1239, USA {liux,tu,wheeler}@research.ge.com

More information

Accurate and robust image superresolution by neural processing of local image representations

Accurate and robust image superresolution by neural processing of local image representations Accurate and robust image superresolution by neural processing of local image representations Carlos Miravet 1,2 and Francisco B. Rodríguez 1 1 Grupo de Neurocomputación Biológica (GNB), Escuela Politécnica

More information

Super-Resolution Imaging : Use of Zoom as a Cue

Super-Resolution Imaging : Use of Zoom as a Cue Super-Resolution Imaging : Use of Zoom as a Cue M.V. Joshi and Subhasis Chaudhuri Department of Electrical Engineering Indian Institute of Technology - Bombay Powai, Mumbai-400076. India mvjoshi, sc@ee.iitb.ac.in

More information

Superresolution images reconstructed from aliased images

Superresolution images reconstructed from aliased images Superresolution images reconstructed from aliased images Patrick Vandewalle, Sabine Süsstrunk and Martin Vetterli LCAV - School of Computer and Communication Sciences Ecole Polytechnique Fédérale de Lausanne

More information

A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow

A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow , pp.233-237 http://dx.doi.org/10.14257/astl.2014.51.53 A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow Giwoo Kim 1, Hye-Youn Lim 1 and Dae-Seong Kang 1, 1 Department of electronices

More information

A Sampled Texture Prior for Image Super-Resolution

A Sampled Texture Prior for Image Super-Resolution A Sampled Texture Prior for Image Super-Resolution Lyndsey C. Pickup, Stephen J. Roberts and Andrew Zisserman Robotics Research Group Department of Engineering Science University of Oxford Parks Road,

More information

Course: Model, Learning, and Inference: Lecture 5

Course: Model, Learning, and Inference: Lecture 5 Course: Model, Learning, and Inference: Lecture 5 Alan Yuille Department of Statistics, UCLA Los Angeles, CA 90095 yuille@stat.ucla.edu Abstract Probability distributions on structured representation.

More information

A PHOTOGRAMMETRIC APPRAOCH FOR AUTOMATIC TRAFFIC ASSESSMENT USING CONVENTIONAL CCTV CAMERA

A PHOTOGRAMMETRIC APPRAOCH FOR AUTOMATIC TRAFFIC ASSESSMENT USING CONVENTIONAL CCTV CAMERA A PHOTOGRAMMETRIC APPRAOCH FOR AUTOMATIC TRAFFIC ASSESSMENT USING CONVENTIONAL CCTV CAMERA N. Zarrinpanjeh a, F. Dadrassjavan b, H. Fattahi c * a Islamic Azad University of Qazvin - nzarrin@qiau.ac.ir

More information

Simultaneous Gamma Correction and Registration in the Frequency Domain

Simultaneous Gamma Correction and Registration in the Frequency Domain Simultaneous Gamma Correction and Registration in the Frequency Domain Alexander Wong a28wong@uwaterloo.ca William Bishop wdbishop@uwaterloo.ca Department of Electrical and Computer Engineering University

More information

Video stabilization for high resolution images reconstruction

Video stabilization for high resolution images reconstruction Advanced Project S9 Video stabilization for high resolution images reconstruction HIMMICH Youssef, KEROUANTON Thomas, PATIES Rémi, VILCHES José. Abstract Super-resolution reconstruction produces one or

More information

CHAPTER 6 TEXTURE ANIMATION

CHAPTER 6 TEXTURE ANIMATION CHAPTER 6 TEXTURE ANIMATION 6.1. INTRODUCTION Animation is the creating of a timed sequence or series of graphic images or frames together to give the appearance of continuous movement. A collection of

More information

The Role of Size Normalization on the Recognition Rate of Handwritten Numerals

The Role of Size Normalization on the Recognition Rate of Handwritten Numerals The Role of Size Normalization on the Recognition Rate of Handwritten Numerals Chun Lei He, Ping Zhang, Jianxiong Dong, Ching Y. Suen, Tien D. Bui Centre for Pattern Recognition and Machine Intelligence,

More information

Image Super-Resolution via Sparse Representation

Image Super-Resolution via Sparse Representation 1 Image Super-Resolution via Sparse Representation Jianchao Yang, Student Member, IEEE, John Wright, Student Member, IEEE Thomas Huang, Life Fellow, IEEE and Yi Ma, Senior Member, IEEE Abstract This paper

More information

Outline. Multitemporal high-resolution image classification

Outline. Multitemporal high-resolution image classification IGARSS-2011 Vancouver, Canada, July 24-29, 29, 2011 Multitemporal Region-Based Classification of High-Resolution Images by Markov Random Fields and Multiscale Segmentation Gabriele Moser Sebastiano B.

More information

Tracking Groups of Pedestrians in Video Sequences

Tracking Groups of Pedestrians in Video Sequences Tracking Groups of Pedestrians in Video Sequences Jorge S. Marques Pedro M. Jorge Arnaldo J. Abrantes J. M. Lemos IST / ISR ISEL / IST ISEL INESC-ID / IST Lisbon, Portugal Lisbon, Portugal Lisbon, Portugal

More information

A NEW SUPER RESOLUTION TECHNIQUE FOR RANGE DATA. Valeria Garro, Pietro Zanuttigh, Guido M. Cortelazzo. University of Padova, Italy

A NEW SUPER RESOLUTION TECHNIQUE FOR RANGE DATA. Valeria Garro, Pietro Zanuttigh, Guido M. Cortelazzo. University of Padova, Italy A NEW SUPER RESOLUTION TECHNIQUE FOR RANGE DATA Valeria Garro, Pietro Zanuttigh, Guido M. Cortelazzo University of Padova, Italy ABSTRACT Current Time-of-Flight matrix sensors allow for the acquisition

More information

NEIGHBORHOOD REGRESSION FOR EDGE-PRESERVING IMAGE SUPER-RESOLUTION. Yanghao Li, Jiaying Liu, Wenhan Yang, Zongming Guo

NEIGHBORHOOD REGRESSION FOR EDGE-PRESERVING IMAGE SUPER-RESOLUTION. Yanghao Li, Jiaying Liu, Wenhan Yang, Zongming Guo NEIGHBORHOOD REGRESSION FOR EDGE-PRESERVING IMAGE SUPER-RESOLUTION Yanghao Li, Jiaying Liu, Wenhan Yang, Zongming Guo Institute of Computer Science and Technology, Peking University, Beijing, P.R.China,

More information

ROBUST COLOR JOINT MULTI-FRAME DEMOSAICING AND SUPER- RESOLUTION ALGORITHM

ROBUST COLOR JOINT MULTI-FRAME DEMOSAICING AND SUPER- RESOLUTION ALGORITHM ROBUST COLOR JOINT MULTI-FRAME DEMOSAICING AND SUPER- RESOLUTION ALGORITHM Theodor Heinze Hasso-Plattner-Institute for Software Systems Engineering Prof.-Dr.-Helmert-Str. 2-3, 14482 Potsdam, Germany theodor.heinze@hpi.uni-potsdam.de

More information

An Application of Inverse Reinforcement Learning to Medical Records of Diabetes Treatment

An Application of Inverse Reinforcement Learning to Medical Records of Diabetes Treatment An Application of Inverse Reinforcement Learning to Medical Records of Diabetes Treatment Hideki Asoh 1, Masanori Shiro 1 Shotaro Akaho 1, Toshihiro Kamishima 1, Koiti Hasida 1, Eiji Aramaki 2, and Takahide

More information

Template-based Eye and Mouth Detection for 3D Video Conferencing

Template-based Eye and Mouth Detection for 3D Video Conferencing Template-based Eye and Mouth Detection for 3D Video Conferencing Jürgen Rurainsky and Peter Eisert Fraunhofer Institute for Telecommunications - Heinrich-Hertz-Institute, Image Processing Department, Einsteinufer

More information

Image Hallucination Using Neighbor Embedding over Visual Primitive Manifolds

Image Hallucination Using Neighbor Embedding over Visual Primitive Manifolds Image Hallucination Using Neighbor Embedding over Visual Primitive Manifolds Wei Fan & Dit-Yan Yeung Department of Computer Science and Engineering, Hong Kong University of Science and Technology {fwkevin,dyyeung}@cse.ust.hk

More information

Probabilistic Latent Semantic Analysis (plsa)

Probabilistic Latent Semantic Analysis (plsa) Probabilistic Latent Semantic Analysis (plsa) SS 2008 Bayesian Networks Multimedia Computing, Universität Augsburg Rainer.Lienhart@informatik.uni-augsburg.de www.multimedia-computing.{de,org} References

More information

Low-resolution Character Recognition by Video-based Super-resolution

Low-resolution Character Recognition by Video-based Super-resolution 2009 10th International Conference on Document Analysis and Recognition Low-resolution Character Recognition by Video-based Super-resolution Ataru Ohkura 1, Daisuke Deguchi 1, Tomokazu Takahashi 2, Ichiro

More information

The Scientific Data Mining Process

The Scientific Data Mining Process Chapter 4 The Scientific Data Mining Process When I use a word, Humpty Dumpty said, in rather a scornful tone, it means just what I choose it to mean neither more nor less. Lewis Carroll [87, p. 214] In

More information

An Iterative Image Registration Technique with an Application to Stereo Vision

An Iterative Image Registration Technique with an Application to Stereo Vision An Iterative Image Registration Technique with an Application to Stereo Vision Bruce D. Lucas Takeo Kanade Computer Science Department Carnegie-Mellon University Pittsburgh, Pennsylvania 15213 Abstract

More information

Super-Resolution for Traditional and Omnidirectional Image Sequences

Super-Resolution for Traditional and Omnidirectional Image Sequences Acta Polytechnica Hungarica Vol. 6, No. 1, 2009 Super-Resolution for Traditional and Omnidirectional Image Sequences Attila Nagy, Zoltán Vámossy Institute of Software Technology, John von Neumann Faculty

More information

MapReduce Approach to Collective Classification for Networks

MapReduce Approach to Collective Classification for Networks MapReduce Approach to Collective Classification for Networks Wojciech Indyk 1, Tomasz Kajdanowicz 1, Przemyslaw Kazienko 1, and Slawomir Plamowski 1 Wroclaw University of Technology, Wroclaw, Poland Faculty

More information

Mean-Shift Tracking with Random Sampling

Mean-Shift Tracking with Random Sampling 1 Mean-Shift Tracking with Random Sampling Alex Po Leung, Shaogang Gong Department of Computer Science Queen Mary, University of London, London, E1 4NS Abstract In this work, boosting the efficiency of

More information

Image super-resolution: Historical overview and future challenges

Image super-resolution: Historical overview and future challenges 1 Image super-resolution: Historical overview and future challenges Jianchao Yang University of Illinois at Urbana-Champaign Thomas Huang University of Illinois at Urbana-Champaign CONTENTS 1.1 Introduction

More information

Edge tracking for motion segmentation and depth ordering

Edge tracking for motion segmentation and depth ordering Edge tracking for motion segmentation and depth ordering P. Smith, T. Drummond and R. Cipolla Department of Engineering University of Cambridge Cambridge CB2 1PZ,UK {pas1001 twd20 cipolla}@eng.cam.ac.uk

More information

Automatic 3D Reconstruction via Object Detection and 3D Transformable Model Matching CS 269 Class Project Report

Automatic 3D Reconstruction via Object Detection and 3D Transformable Model Matching CS 269 Class Project Report Automatic 3D Reconstruction via Object Detection and 3D Transformable Model Matching CS 69 Class Project Report Junhua Mao and Lunbo Xu University of California, Los Angeles mjhustc@ucla.edu and lunbo

More information

Subspace Analysis and Optimization for AAM Based Face Alignment

Subspace Analysis and Optimization for AAM Based Face Alignment Subspace Analysis and Optimization for AAM Based Face Alignment Ming Zhao Chun Chen College of Computer Science Zhejiang University Hangzhou, 310027, P.R.China zhaoming1999@zju.edu.cn Stan Z. Li Microsoft

More information

Image Super-Resolution as Sparse Representation of Raw Image Patches

Image Super-Resolution as Sparse Representation of Raw Image Patches Image Super-Resolution as Sparse Representation of Raw Image Patches Jianchao Yang, John Wright, Yi Ma, Thomas Huang University of Illinois at Urbana-Champagin Beckman Institute and Coordinated Science

More information

Fast Matching of Binary Features

Fast Matching of Binary Features Fast Matching of Binary Features Marius Muja and David G. Lowe Laboratory for Computational Intelligence University of British Columbia, Vancouver, Canada {mariusm,lowe}@cs.ubc.ca Abstract There has been

More information

Performance Verification of Super-Resolution Image Reconstruction

Performance Verification of Super-Resolution Image Reconstruction Performance Verification of Super-Resolution Image Reconstruction Masaki Sugie Department of Information Science, Kogakuin University Tokyo, Japan Email: em13010@ns.kogakuin.ac.jp Seiichi Gohshi Department

More information

Latest Results on High-Resolution Reconstruction from Video Sequences

Latest Results on High-Resolution Reconstruction from Video Sequences Latest Results on High-Resolution Reconstruction from Video Sequences S. Lertrattanapanich and N. K. Bose The Spatial and Temporal Signal Processing Center Department of Electrical Engineering The Pennsylvania

More information

Determining optimal window size for texture feature extraction methods

Determining optimal window size for texture feature extraction methods IX Spanish Symposium on Pattern Recognition and Image Analysis, Castellon, Spain, May 2001, vol.2, 237-242, ISBN: 84-8021-351-5. Determining optimal window size for texture feature extraction methods Domènec

More information

Low-resolution Image Processing based on FPGA

Low-resolution Image Processing based on FPGA Abstract Research Journal of Recent Sciences ISSN 2277-2502. Low-resolution Image Processing based on FPGA Mahshid Aghania Kiau, Islamic Azad university of Karaj, IRAN Available online at: www.isca.in,

More information

Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches

Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches PhD Thesis by Payam Birjandi Director: Prof. Mihai Datcu Problematic

More information

Super-resolution Reconstruction Algorithm Based on Patch Similarity and Back-projection Modification

Super-resolution Reconstruction Algorithm Based on Patch Similarity and Back-projection Modification 1862 JOURNAL OF SOFTWARE, VOL 9, NO 7, JULY 214 Super-resolution Reconstruction Algorithm Based on Patch Similarity and Back-projection Modification Wei-long Chen Digital Media College, Sichuan Normal

More information

A Novel Method to Improve Resolution of Satellite Images Using DWT and Interpolation

A Novel Method to Improve Resolution of Satellite Images Using DWT and Interpolation A Novel Method to Improve Resolution of Satellite Images Using DWT and Interpolation S.VENKATA RAMANA ¹, S. NARAYANA REDDY ² M.Tech student, Department of ECE, SVU college of Engineering, Tirupati, 517502,

More information

Machine Learning and Data Analysis overview. Department of Cybernetics, Czech Technical University in Prague. http://ida.felk.cvut.

Machine Learning and Data Analysis overview. Department of Cybernetics, Czech Technical University in Prague. http://ida.felk.cvut. Machine Learning and Data Analysis overview Jiří Kléma Department of Cybernetics, Czech Technical University in Prague http://ida.felk.cvut.cz psyllabus Lecture Lecturer Content 1. J. Kléma Introduction,

More information

Spatial Accuracy Assessment of Digital Surface Models: A Probabilistic Approach

Spatial Accuracy Assessment of Digital Surface Models: A Probabilistic Approach Spatial Accuracy Assessment of Digital Surface Models: A Probabilistic Approach André Jalobeanu Centro de Geofisica de Evora, Portugal part of the SPACEFUSION Project (ANR, France) Outline Goals and objectives

More information

FAST REGISTRATION METHODS FOR SUPER-RESOLUTION IMAGING. Jari Hannuksela, Jarno Väyrynen, Janne Heikkilä and Pekka Sangi

FAST REGISTRATION METHODS FOR SUPER-RESOLUTION IMAGING. Jari Hannuksela, Jarno Väyrynen, Janne Heikkilä and Pekka Sangi FAST REGISTRATION METHODS FOR SUPER-RESOLUTION IMAGING Jari Hannuksela, Jarno Väyrynen, Janne Heikkilä and Pekka Sangi Machine Vision Group, Infotech Oulu Department of Electrical and Information Engineering

More information

A System for Capturing High Resolution Images

A System for Capturing High Resolution Images A System for Capturing High Resolution Images G.Voyatzis, G.Angelopoulos, A.Bors and I.Pitas Department of Informatics University of Thessaloniki BOX 451, 54006 Thessaloniki GREECE e-mail: pitas@zeus.csd.auth.gr

More information

ROBUST VEHICLE TRACKING IN VIDEO IMAGES BEING TAKEN FROM A HELICOPTER

ROBUST VEHICLE TRACKING IN VIDEO IMAGES BEING TAKEN FROM A HELICOPTER ROBUST VEHICLE TRACKING IN VIDEO IMAGES BEING TAKEN FROM A HELICOPTER Fatemeh Karimi Nejadasl, Ben G.H. Gorte, and Serge P. Hoogendoorn Institute of Earth Observation and Space System, Delft University

More information

LABELING IN VIDEO DATABASES. School of Electrical and Computer Engineering. 1285 Electrical Engineering Building. in the video database.

LABELING IN VIDEO DATABASES. School of Electrical and Computer Engineering. 1285 Electrical Engineering Building. in the video database. FACE DETECTION FOR PSEUDO-SEMANTIC LABELING IN VIDEO DATABASES Alberto Albiol Departamento de Comunicaciones Universidad Politçecnica de Valencia Valencia, Spain email: alalbiol@dcom.upv.es Charles A.

More information

Statistical Machine Learning from Data

Statistical Machine Learning from Data Samy Bengio Statistical Machine Learning from Data 1 Statistical Machine Learning from Data Gaussian Mixture Models Samy Bengio IDIAP Research Institute, Martigny, Switzerland, and Ecole Polytechnique

More information

LABEL PROPAGATION ON GRAPHS. SEMI-SUPERVISED LEARNING. ----Changsheng Liu 10-30-2014

LABEL PROPAGATION ON GRAPHS. SEMI-SUPERVISED LEARNING. ----Changsheng Liu 10-30-2014 LABEL PROPAGATION ON GRAPHS. SEMI-SUPERVISED LEARNING ----Changsheng Liu 10-30-2014 Agenda Semi Supervised Learning Topics in Semi Supervised Learning Label Propagation Local and global consistency Graph

More information

Machine Learning in Multi-frame Image Super-resolution

Machine Learning in Multi-frame Image Super-resolution Machine Learning in Multi-frame Image Super-resolution Lyndsey C. Pickup Robotics Research Group Department of Engineering Science University of Oxford Michaelmas Term, 2007 Lyndsey C. Pickup Doctor of

More information

Using Data Mining for Mobile Communication Clustering and Characterization

Using Data Mining for Mobile Communication Clustering and Characterization Using Data Mining for Mobile Communication Clustering and Characterization A. Bascacov *, C. Cernazanu ** and M. Marcu ** * Lasting Software, Timisoara, Romania ** Politehnica University of Timisoara/Computer

More information

PHYSIOLOGICALLY-BASED DETECTION OF COMPUTER GENERATED FACES IN VIDEO

PHYSIOLOGICALLY-BASED DETECTION OF COMPUTER GENERATED FACES IN VIDEO PHYSIOLOGICALLY-BASED DETECTION OF COMPUTER GENERATED FACES IN VIDEO V. Conotter, E. Bodnari, G. Boato H. Farid Department of Information Engineering and Computer Science University of Trento, Trento (ITALY)

More information

Two-Frame Motion Estimation Based on Polynomial Expansion

Two-Frame Motion Estimation Based on Polynomial Expansion Two-Frame Motion Estimation Based on Polynomial Expansion Gunnar Farnebäck Computer Vision Laboratory, Linköping University, SE-581 83 Linköping, Sweden gf@isy.liu.se http://www.isy.liu.se/cvl/ Abstract.

More information

Circle Object Recognition Based on Monocular Vision for Home Security Robot

Circle Object Recognition Based on Monocular Vision for Home Security Robot Journal of Applied Science and Engineering, Vol. 16, No. 3, pp. 261 268 (2013) DOI: 10.6180/jase.2013.16.3.05 Circle Object Recognition Based on Monocular Vision for Home Security Robot Shih-An Li, Ching-Chang

More information

Face Recognition in Low-resolution Images by Using Local Zernike Moments

Face Recognition in Low-resolution Images by Using Local Zernike Moments Proceedings of the International Conference on Machine Vision and Machine Learning Prague, Czech Republic, August14-15, 014 Paper No. 15 Face Recognition in Low-resolution Images by Using Local Zernie

More information

RESOLUTION IMPROVEMENT OF DIGITIZED IMAGES

RESOLUTION IMPROVEMENT OF DIGITIZED IMAGES Proceedings of ALGORITMY 2005 pp. 270 279 RESOLUTION IMPROVEMENT OF DIGITIZED IMAGES LIBOR VÁŠA AND VÁCLAV SKALA Abstract. A quick overview of preprocessing performed by digital still cameras is given

More information

Linear Threshold Units

Linear Threshold Units Linear Threshold Units w x hx (... w n x n w We assume that each feature x j and each weight w j is a real number (we will relax this later) We will study three different algorithms for learning linear

More information

PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 4: LINEAR MODELS FOR CLASSIFICATION

PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 4: LINEAR MODELS FOR CLASSIFICATION PATTERN RECOGNITION AND MACHINE LEARNING CHAPTER 4: LINEAR MODELS FOR CLASSIFICATION Introduction In the previous chapter, we explored a class of regression models having particularly simple analytical

More information

NEURAL NETWORKS A Comprehensive Foundation

NEURAL NETWORKS A Comprehensive Foundation NEURAL NETWORKS A Comprehensive Foundation Second Edition Simon Haykin McMaster University Hamilton, Ontario, Canada Prentice Hall Prentice Hall Upper Saddle River; New Jersey 07458 Preface xii Acknowledgments

More information

Christfried Webers. Canberra February June 2015

Christfried Webers. Canberra February June 2015 c Statistical Group and College of Engineering and Computer Science Canberra February June (Many figures from C. M. Bishop, "Pattern Recognition and ") 1of 829 c Part VIII Linear Classification 2 Logistic

More information

A Reliability Point and Kalman Filter-based Vehicle Tracking Technique

A Reliability Point and Kalman Filter-based Vehicle Tracking Technique A Reliability Point and Kalman Filter-based Vehicle Tracing Technique Soo Siang Teoh and Thomas Bräunl Abstract This paper introduces a technique for tracing the movement of vehicles in consecutive video

More information

Character Image Patterns as Big Data

Character Image Patterns as Big Data 22 International Conference on Frontiers in Handwriting Recognition Character Image Patterns as Big Data Seiichi Uchida, Ryosuke Ishida, Akira Yoshida, Wenjie Cai, Yaokai Feng Kyushu University, Fukuoka,

More information

Visualization of Breast Cancer Data by SOM Component Planes

Visualization of Breast Cancer Data by SOM Component Planes International Journal of Science and Technology Volume 3 No. 2, February, 2014 Visualization of Breast Cancer Data by SOM Component Planes P.Venkatesan. 1, M.Mullai 2 1 Department of Statistics,NIRT(Indian

More information

MACHINE VISION MNEMONICS, INC. 102 Gaither Drive, Suite 4 Mount Laurel, NJ 08054 USA 856-234-0970 www.mnemonicsinc.com

MACHINE VISION MNEMONICS, INC. 102 Gaither Drive, Suite 4 Mount Laurel, NJ 08054 USA 856-234-0970 www.mnemonicsinc.com MACHINE VISION by MNEMONICS, INC. 102 Gaither Drive, Suite 4 Mount Laurel, NJ 08054 USA 856-234-0970 www.mnemonicsinc.com Overview A visual information processing company with over 25 years experience

More information

Object tracking & Motion detection in video sequences

Object tracking & Motion detection in video sequences Introduction Object tracking & Motion detection in video sequences Recomended link: http://cmp.felk.cvut.cz/~hlavac/teachpresen/17compvision3d/41imagemotion.pdf 1 2 DYNAMIC SCENE ANALYSIS The input to

More information

CS 688 Pattern Recognition Lecture 4. Linear Models for Classification

CS 688 Pattern Recognition Lecture 4. Linear Models for Classification CS 688 Pattern Recognition Lecture 4 Linear Models for Classification Probabilistic generative models Probabilistic discriminative models 1 Generative Approach ( x ) p C k p( C k ) Ck p ( ) ( x Ck ) p(

More information

Norbert Schuff Professor of Radiology VA Medical Center and UCSF Norbert.schuff@ucsf.edu

Norbert Schuff Professor of Radiology VA Medical Center and UCSF Norbert.schuff@ucsf.edu Norbert Schuff Professor of Radiology Medical Center and UCSF Norbert.schuff@ucsf.edu Medical Imaging Informatics 2012, N.Schuff Course # 170.03 Slide 1/67 Overview Definitions Role of Segmentation Segmentation

More information

Super-resolution method based on edge feature for high resolution imaging

Super-resolution method based on edge feature for high resolution imaging Science Journal of Circuits, Systems and Signal Processing 2014; 3(6-1): 24-29 Published online December 26, 2014 (http://www.sciencepublishinggroup.com/j/cssp) doi: 10.11648/j.cssp.s.2014030601.14 ISSN:

More information

Vision based Vehicle Tracking using a high angle camera

Vision based Vehicle Tracking using a high angle camera Vision based Vehicle Tracking using a high angle camera Raúl Ignacio Ramos García Dule Shu gramos@clemson.edu dshu@clemson.edu Abstract A vehicle tracking and grouping algorithm is presented in this work

More information

How To Segmentate In Ctv Video

How To Segmentate In Ctv Video Time and Date OCR in CCTV Video Ginés García-Mateos 1, Andrés García-Meroño 1, Cristina Vicente-Chicote 3, Alberto Ruiz 1, and Pedro E. López-de-Teruel 2 1 Dept. de Informática y Sistemas 2 Dept. de Ingeniería

More information

2.11 CMC curves showing the performance of sketch to digital face image matching. algorithms on the CUHK database... 40

2.11 CMC curves showing the performance of sketch to digital face image matching. algorithms on the CUHK database... 40 List of Figures 1.1 Illustrating different stages in a face recognition system i.e. image acquisition, face detection, face normalization, feature extraction, and matching.. 10 1.2 Illustrating the concepts

More information

Signature Segmentation from Machine Printed Documents using Conditional Random Field

Signature Segmentation from Machine Printed Documents using Conditional Random Field 2011 International Conference on Document Analysis and Recognition Signature Segmentation from Machine Printed Documents using Conditional Random Field Ranju Mandal Computer Vision and Pattern Recognition

More information

3D Model based Object Class Detection in An Arbitrary View

3D Model based Object Class Detection in An Arbitrary View 3D Model based Object Class Detection in An Arbitrary View Pingkun Yan, Saad M. Khan, Mubarak Shah School of Electrical Engineering and Computer Science University of Central Florida http://www.eecs.ucf.edu/

More information

Robert Collins CSE598G. More on Mean-shift. R.Collins, CSE, PSU CSE598G Spring 2006

Robert Collins CSE598G. More on Mean-shift. R.Collins, CSE, PSU CSE598G Spring 2006 More on Mean-shift R.Collins, CSE, PSU Spring 2006 Recall: Kernel Density Estimation Given a set of data samples x i ; i=1...n Convolve with a kernel function H to generate a smooth function f(x) Equivalent

More information

Combining High-Resolution Images With Low-Quality Videos

Combining High-Resolution Images With Low-Quality Videos Combining High-Resolution Images With Low-Quality Videos Falk Schubert EADS Innovation Works, Germany falk.schubert@eads.net Krystian Mikolajczyk University of Surrey, UK k.mikolajczyk@surrey.ac.uk Abstract

More information

Automatic parameter regulation for a tracking system with an auto-critical function

Automatic parameter regulation for a tracking system with an auto-critical function Automatic parameter regulation for a tracking system with an auto-critical function Daniela Hall INRIA Rhône-Alpes, St. Ismier, France Email: Daniela.Hall@inrialpes.fr Abstract In this article we propose

More information

Probabilistic Models for Big Data. Alex Davies and Roger Frigola University of Cambridge 13th February 2014

Probabilistic Models for Big Data. Alex Davies and Roger Frigola University of Cambridge 13th February 2014 Probabilistic Models for Big Data Alex Davies and Roger Frigola University of Cambridge 13th February 2014 The State of Big Data Why probabilistic models for Big Data? 1. If you don t have to worry about

More information

High Quality Image Deblurring Panchromatic Pixels

High Quality Image Deblurring Panchromatic Pixels High Quality Image Deblurring Panchromatic Pixels ACM Transaction on Graphics vol. 31, No. 5, 2012 Sen Wang, Tingbo Hou, John Border, Hong Qin, and Rodney Miller Presented by Bong-Seok Choi School of Electrical

More information

High Resolution Images from Low Resolution Video Sequences

High Resolution Images from Low Resolution Video Sequences High Resolution Images from Low Resolution Video Sequences Federico Cristina fcristina@lidi.info.unlp.edu.ar - Ayudante Diplomado UNLP Sebastián Dapoto sdapoto@lidi.info.unlp.edu.ar - Becario III-LIDI

More information

The Basics of Graphical Models

The Basics of Graphical Models The Basics of Graphical Models David M. Blei Columbia University October 3, 2015 Introduction These notes follow Chapter 2 of An Introduction to Probabilistic Graphical Models by Michael Jordan. Many figures

More information

How Conditional Random Fields Learn Dynamics: An Example-Based Study

How Conditional Random Fields Learn Dynamics: An Example-Based Study Computer Communication & Collaboration (2013) Submitted on 27/May/2013 How Conditional Random Fields Learn Dynamics: An Example-Based Study Mohammad Javad Shafiee School of Electrical & Computer Engineering,

More information

Redundant Wavelet Transform Based Image Super Resolution

Redundant Wavelet Transform Based Image Super Resolution Redundant Wavelet Transform Based Image Super Resolution Arti Sharma, Prof. Preety D Swami Department of Electronics &Telecommunication Samrat Ashok Technological Institute Vidisha Department of Electronics

More information

Local features and matching. Image classification & object localization

Local features and matching. Image classification & object localization Overview Instance level search Local features and matching Efficient visual recognition Image classification & object localization Category recognition Image classification: assigning a class label to

More information

Colour Image Segmentation Technique for Screen Printing

Colour Image Segmentation Technique for Screen Printing 60 R.U. Hewage and D.U.J. Sonnadara Department of Physics, University of Colombo, Sri Lanka ABSTRACT Screen-printing is an industry with a large number of applications ranging from printing mobile phone

More information

Vector storage and access; algorithms in GIS. This is lecture 6

Vector storage and access; algorithms in GIS. This is lecture 6 Vector storage and access; algorithms in GIS This is lecture 6 Vector data storage and access Vectors are built from points, line and areas. (x,y) Surface: (x,y,z) Vector data access Access to vector

More information

Introduction to Deep Learning Variational Inference, Mean Field Theory

Introduction to Deep Learning Variational Inference, Mean Field Theory Introduction to Deep Learning Variational Inference, Mean Field Theory 1 Iasonas Kokkinos Iasonas.kokkinos@ecp.fr Center for Visual Computing Ecole Centrale Paris Galen Group INRIA-Saclay Lecture 3: recap

More information

Prediction of Heart Disease Using Naïve Bayes Algorithm

Prediction of Heart Disease Using Naïve Bayes Algorithm Prediction of Heart Disease Using Naïve Bayes Algorithm R.Karthiyayini 1, S.Chithaara 2 Assistant Professor, Department of computer Applications, Anna University, BIT campus, Tiruchirapalli, Tamilnadu,

More information

International Journal of Computer Science Trends and Technology (IJCST) Volume 2 Issue 3, May-Jun 2014

International Journal of Computer Science Trends and Technology (IJCST) Volume 2 Issue 3, May-Jun 2014 RESEARCH ARTICLE OPEN ACCESS A Survey of Data Mining: Concepts with Applications and its Future Scope Dr. Zubair Khan 1, Ashish Kumar 2, Sunny Kumar 3 M.Tech Research Scholar 2. Department of Computer

More information

Understanding Purposeful Human Motion

Understanding Purposeful Human Motion M.I.T Media Laboratory Perceptual Computing Section Technical Report No. 85 Appears in Fourth IEEE International Conference on Automatic Face and Gesture Recognition Understanding Purposeful Human Motion

More information

An Active Head Tracking System for Distance Education and Videoconferencing Applications

An Active Head Tracking System for Distance Education and Videoconferencing Applications An Active Head Tracking System for Distance Education and Videoconferencing Applications Sami Huttunen and Janne Heikkilä Machine Vision Group Infotech Oulu and Department of Electrical and Information

More information

Automatic Restoration Algorithms for 35mm film

Automatic Restoration Algorithms for 35mm film P. Schallauer, A. Pinz, W. Haas. Automatic Restoration Algorithms for 35mm film. To be published in Videre, Journal of Computer Vision Research, web: http://mitpress.mit.edu/videre.html, 1999. Automatic

More information

A Statistical Framework for Operational Infrasound Monitoring

A Statistical Framework for Operational Infrasound Monitoring A Statistical Framework for Operational Infrasound Monitoring Stephen J. Arrowsmith Rod W. Whitaker LA-UR 11-03040 The views expressed here do not necessarily reflect the views of the United States Government,

More information

Statistical Machine Learning

Statistical Machine Learning Statistical Machine Learning UoC Stats 37700, Winter quarter Lecture 4: classical linear and quadratic discriminants. 1 / 25 Linear separation For two classes in R d : simple idea: separate the classes

More information