Papers by Georgy Gimelfarb
Lecture Notes in Computer Science, 2006
This paper addresses the problem of facial expressions recognition using principal component anal... more This paper addresses the problem of facial expressions recognition using principal component analysis and independent component analysis onto dimension of the emotion. To reflect well the changes in facial expressions, a representation based on principal component analysis (PCA) excluded the first 2 principal components is presented, ICA representation from this PCA representation is developed. Facial expression performance in two dimensional structure was significant 90.9% in pleasure/displeasure dimension and 66.6% in the arousal/sleep dimension. The findings indicate that the two dimensional structure of emotion may reflect various emotion states as a stabled structure for the facial expression recognition.
Structural, Syntactic, and Statistical Pattern Recognition
Lecture Notes in Computer Science, 2010
Invited Talks.- From Region Based Image Representation to Object Discovery and Recognition.- Lear... more Invited Talks.- From Region Based Image Representation to Object Discovery and Recognition.- Learning on Manifolds.- Classification and Trees.- Structural Patterns in Complex Networks through Spectral Analysis.- Structural Descriptions.- Graph Embedding Using an Edge-Based Wave Kernel.- A Structured Learning Approach to Attributed Graph Embedding.- Machine Learning.- Combining Elimination Rules in Tree-Based Nearest Neighbor Search Algorithms.- Localized Projection Learning.- Entropy-Based Variational Scheme for Fast Bayes Learning of Gaussian Mixtures.- Structural Learning.- Learning Graph Quantization.- High-Dimensional Spectral Feature Selection for 3D Object Recognition Based on Reeb Graphs.- Dissimilarity-Based Multiple Instance Learning.- A Game Theoretic Approach to Learning Shape Categories and Contextual Similarities.- Poster Session.- A Comparison between Two Representatives of a Set of Graphs: Median vs. Barycenter Graph.- Impact of Visual Information on Text and Content Based Image Retrieval.- Automatic Traffic Monitoring from Satellite Images Using Artificial Immune System.- Graduated Assignment Algorithm for Finding the Common Labelling of a Set of Graphs.- Affinity Propagation for Class Exemplar Mining.- Guided Informative Image Partitioning.- Visual Alphabets on Different Levels of Abstraction for the Recognition of Deformable Objects.- Graph Embedding Based on Nodes Attributes Representatives and a Graph of Words Representation.- Extracting Plane Graphs from Images.- Indexing Tree and Subtree by Using a Structure Network.- Attributed Graph Matching for Image-Features Association Using SIFT Descriptors.- A Causal Extraction Scheme in Top-Down Pyramids for Large Images Segmentation.- Fast Population Game Dynamics for Dominant Sets and Other Quadratic Optimization Problems.- What Is the Complexity of a Network? The Heat Flow-Thermodynamic Depth Approach.- New Partially Labelled Tree Similarity Measure: A Case Study.- Complete Search Space Exploration for SITG Inside Probability.- Commute-Time Convolution Kernels for Graph Clustering.- Geometric Methods.- Non-Euclidean Dissimilarities: Causes and Informativeness.- Non-parametric Mixture Models for Clustering.- Structural Methods for Vision.- A Probabilistic Approach to Spectral Unmixing.- A Game-Theoretic Approach to the Enforcement of Global Consistency in Multi-view Feature Matching.- An Algorithm for Recovering Camouflage Errors on Moving People.- Clustering.- Semi-supervised Clustering Using Heterogeneous Dissimilarities.- On Consensus Clustering Validation.- Pairwise Probabilistic Clustering Using Evidence Accumulation.- Exploring the Performance Limit of Cluster Ensemble Techniques.- Contour Grouping by Clustering with Multi-feature Similarity Measure.- Poster Session.- A Psychophysical Evaluation of Texture Degradation Descriptors.- Content-Based Tile Retrieval System.- Performance Improvement in Multiple-Model Speech Recognizer under Noisy Environments.- On Feature Combination for Music Classification.- Information Theoretical Kernels for Generative Embeddings Based on Hidden Markov Models.- Dynamic Linear Combination of Two-Class Classifiers.- Large-Scale Text to Image Retrieval Using a Bayesian K-Neighborhood Model.- Maximum a Posteriori Based Kernel Classifier Trained by Linear Programming.- Improvement of the Disc Harmonic Moments Descriptor by an Exponentially Decaying Distance Transform.- Feature Level Fusion of Face and Palmprint Biometrics.- Scale and Rotation Invariant Detection of Singular Patterns in Vector Flow Fields.- Using K-NN SVMs for Performance Improvement and Comparison to K-Highest Lagrange Multipliers Selection.- Automatic Speech Segmentation Based on Acoustical Clustering.- An Efficient Iris and Eye Corners Extraction Method.- Dissimilarity-Based Methods.- An Empirical Comparison of Kernel-Based and Dissimilarity-Based Feature Spaces.- The Dissimilarity Representation as a Tool for Three-Way Data Classification: A 2D Measure.- Regularising the Ricci Flow Embedding.- Spherical Embedding and Classification.- Language.- Language Detection and Tracking in Multilingual Documents Using Weak Estimators.- Similarity Word-Sequence Kernels for Sentence Clustering.- Bayesian Adaptation for Statistical Machine Translation.- A Generative Score Space for Statistical Dialog Characterization in Social Signalling.- Multiple Classifiers.- A Modular Approach to Training Cascades of Boosted Ensembles.- A Linear Combination of Classifiers via Rank Margin Maximization.- Combination of Dichotomizers for Maximizing the Partial Area under the ROC Curve.- Graphs.- Ihara Coefficients: A Flexible Tool for Higher Order Learning.- A New Spectral Bound on the Clique Number of Graphs.- Large Sample Statistics in the Domain of Graphs.- Statistical Pattern Recognition.- Analysis of the Multi-Dimensional Scale Saliency Algorithm and Its Application to Texture Categorization.- Interactive Image Retrieval Using Smoothed Nearest Neighbor Estimates.-…
A Novel Autoencoder-Based Diagnostic System for Early Assessment of Lung Cancer
2018 25th IEEE International Conference on Image Processing (ICIP), 2018
A novel framework for the classification of lung nodules using computed tomography (CT) scans is ... more A novel framework for the classification of lung nodules using computed tomography (CT) scans is proposed in this paper. To get an accurate diagnosis of the detected lung nodules, the proposed framework integrates the following two groups of features: (i) appearance features that is modeled using higher-order Markov Gibbs random field (MGRF)-model that has the ability to describe the spatial inhomogeneities inside the lung nodule; and (ii) geometric features that describe the shape geometry of the lung nodules. The novelty of this paper is to accurately model the appearance of the detected lung nodules using a new developed 7th-order MGRF model that has the ability to model the existing spatial inhomogeneities for both small and large detected lung nodules, in addition to the integration with the extracted geometric features. Finally, a deep autoencoder (AE) classifier is fed by the above two feature groups to distinguish between the malignant and benign nodules. To evaluate the proposed framework, we used the publicly available data from the Lung Image Database Consortium (LIDC). We used a total of 727 nodules that were collected from 467 patients. The proposed system demonstrates the promise to be a valuable tool for the detection of lung cancer evidenced by achieving a nodule classification accuracy of 92.20%.
2 LCDG-Model of a Multi-modal TOF-MRA Image
We present a fast algorithm for automatic extraction of a 3D cerebrovascular system from time-of-... more We present a fast algorithm for automatic extraction of a 3D cerebrovascular system from time-of-flight (TOF) magnetic resonance angiography (MRA) data. Blood vessels are separated from background tissues (fat, bones, or grey and white brain matter) by voxel-wise classification based on precise approximation of a multi-modal empirical marginal intensity distribution of the TOF-MRA data. The approximation involves a linear combination of discrete Gaussians (LCDG) with alternating signs, and we modify the conventional ExpectationMaximization (EM) algorithm to deal with the LCDG. To validate the accuracy of our algorithm, a special 3D geometrical phantom motivated by statistical analysis of the MRA-TOF data is designed. Experiments with both the phantom and 50 real data sets confirm high accuracy of the proposed approach.
Procedings of the British Machine Vision Conference 2008, 2008
New techniques for more accurate unsupervised segmentation of lung tissues from Low Dose Computed... more New techniques for more accurate unsupervised segmentation of lung tissues from Low Dose Computed Tomography (LDCT) are proposed. In this paper we describe LDCT images and desired maps of regions (lung and the other chest tissues) by a joint Markov-Gibbs random field model (MGRF) of independent image signals and interdependent region labels but focus on most accurate model identification. To better specify region borders, each empirical distribution of signals is precisely approximated by a Linear Combination of Discrete Gaussians (LCDG) with positive and negative components. We modify a conventional Expectation-Maximization (EM) algorithm to deal with the LCDG and develop a sequential EM-based technique to get an initial LCDG-approximation for the modified EM algorithm. The initial segmentation based on the LCDG-models is then iteratively refined using a MGRF model with analytically estimated potentials. Experiments on real data sets confirm high accuracy of the proposed approach.
Unsupervised illness recognition via in-home monitoring by depth cameras
Pervasive and Mobile Computing, 2017
Abstract Most of today’s in-home systems for detecting illness in the elderly meet with the same ... more Abstract Most of today’s in-home systems for detecting illness in the elderly meet with the same limitations: they target only single occupancy apartments, although couples may need support, too, and, faced with monitored subjects who are not necessarily regular in their everyday life, they tend to classify even slight deviations from standard routines (e.g. cooking a meal one hour later than usual) as anomalies. The training data used for anomaly detectors typically include only days when the subject is alone and not sick, whereas in practice it is difficult to obtain day labels, i.e. information on whether the elderly subject had visitors or felt well or unwell. In addition, as every room in the apartment to be monitored is usually equipped with passive infrared motion detectors, the maintaining of such systems may be inconvenient for the residents. To address these problems, this paper proposes a new probabilistic illness detector which does not rely on regular daily routines and requires no data labelling. The detector was tested in apartments inhabited by single elderly subjects or couples, and in all cases only the living rooms and corridors were monitored, with no invasion into the more private spaces. Despite its fully unsupervised training on data covering both normal and unusual days (days of illness, visits by other people, etc.), the proposed detector distinguished between normal days and illnesses with an average accuracy of 88% and did not misclassify the receptions of guests as anomalies.
Image-Based Computer-Aided Diagnostic System for Early Diagnosis of Prostate Cancer
Lecture Notes in Computer Science, 2016
The goal of this paper is to develop a computer-aided diagnostic (CAD) system for early detection... more The goal of this paper is to develop a computer-aided diagnostic (CAD) system for early detection of prostate cancer from diffusion-weighted magnetic resonance imaging (DW-MRI) acquired at different b-values. The proposed system consists of three main steps. First, the prostate is segmented using a hybrid framework that integrates geometric deformable model (level-sets) and nonnegative matrix factorization (NMF). Secondly, the apparent diffusion coefficient (ADC) of the segmented prostate volume is first estimated at different b-values and is then normalized and refined using a generalized Gauss-Markov random field (GGMRF) image model. Then, the cumulative distribution function (CDF) of the refined ADCs at different b-values are constructed. Finally, a two-stage structure of stacked non-negativity constraint auto-encoder (SNCAE) is trained to classify the prostate tumor as benign or malignant based on the constructed CDFs. In the first stage, classification probabilities are estimated at each b-value and in the second stage, those probabilities are fused and fed into the prediction stage SNCAE to calculate the final classification. Preliminary experiments on 53 clinical DW-MRI datasets resulted in \(98.11\,\%\) correct classification (sensitivity \(=96.15\,\%\) and specificity = \(100\,\%\)), indicating the high performance of the proposed CAD system and holding promise of the proposed system as a reliable non-invasive diagnostic tool.
Lecture Notes in Computer Science, 1995
A novel probabilistic model of noisy piecewise-constant images is used to segment real images bei... more A novel probabilistic model of noisy piecewise-constant images is used to segment real images being of interest for ecological monitoring. The model considers a pair composed of a greyseale (or multiband) image and a map of its homogeneous regions as a sample of a Markov random field (MRF) specified by a joint Gibbs probability distribution (GPD) of images and maps. Parameters of the model are estimated by using a stochastic approximation technique. Here, its convergence to the desired values is studied experimentally. Maximum posterior marginal probabilities of region labels for a compound Bayesian segmentation are estimated by generating, with a stochastic relaxation, one or several Markov chains of the region maps under the given GPD. We compare two estimates: (0 traditional sample frequencies and (it') averaged transition probabilities of the labels in each pixel. Experiments in generating the pairs of the images and region maps that correspond to the given joint MRF/GPD model and in segmenting the simulated and real images are discussed.
Detection and recognition of lung nodules in spiral ct images using deformable templates and bayesian post-classification
2004 International Conference on Image Processing, 2004. ICIP '04.
Abstract In this paper, we propose a novel algorithm for isolating lung abnormalities (nodules) f... more Abstract In this paper, we propose a novel algorithm for isolating lung abnormalities (nodules) from low dose spiral chest CT scans. The proposed algorithm consists of three main steps. The first step isolates the lung nodules, arteries, veins, bronchi, and bronchioles from the surrounding anatomical structures. The second step detects lung nodules using deformable 2D and 3D templates describing typical geometry and gray level distribution within the nodules of the same type. The detection combines the normalized cross- ...
Supervised segmentation by pairwise interactions: do Gibbs models learn what we expect?
Proceedings. Fourteenth International Conference on Pattern Recognition (Cat. No.98EX170)
Abstract Gibbs random field image models with multiple translation invariant pairwise pixel inter... more Abstract Gibbs random field image models with multiple translation invariant pairwise pixel interactions show promise for segmenting piecewise-homogeneous image textures because they allow learning of both the interaction structure and strengths from a given training sample. We discuss whether the learnt parameters fit our expectations with respect to discriminating the given textures. Experiments with natural textures show that the learning tends to adapt the model more to peculiarities of the training sample than to general ...
Symmetric Bi- and Trinocular Stereo: Tradeoffs between Theoretical Foundations and Heuristics
Computing Supplement, 1996
R��sum��/Abstract Tradeoffs between theoretical and heuristic sides of ill-posed problems of the ... more R��sum��/Abstract Tradeoffs between theoretical and heuristic sides of ill-posed problems of the intensity-based computational stereo are discussed, as applied to the previously proposed symmetric approach for solving this problem. The heuristics are needed to deal with discontinuities in stereo images due to partial occlusions of observed surface. Basically, it is these discontinuities that cause the ill-posedness of the stereo problems. Theoretical base of the symmetric stereo is refined here by introducing a novel probabilistic model of ...
Procedings of the British Machine Vision Conference 2005, 2005
Deformable or active contour, and surface models are powerful image segmentation techniques. We i... more Deformable or active contour, and surface models are powerful image segmentation techniques. We introduce a novel fast and robust bi-directional parametric deformable model which is able to segment regions of intricate shape in multi-modal greyscale images. The power of the algorithm in terms of computation time and robustness is owing to the use of joint probabilities of the signals and region labels in individual points as external forces guiding the model evolution. These joint probabilities are derived from a Markov-Gibbs random field (MGRF) image model considering an image as a sample of two interrelated spatial stochastic processes. The low level process with conditionally independent and arbitrarily distributed signals relates to the observed image whereas its hidden map of regions is represented with the high level MGRF of interdependent region labels. Marginal probability distributions of signals in each region are recovered from a mixed empirical signal distribution over the whole image. In so doing, each marginal is approximated with a linear combination of Gaussians (LCG) having both positive and negative components. The LCG parameters are estimated using our previously proposed modification of the EM algorithm, and the high-level Gibbs potentials are computed analytically. Comparative experiments show that the proposed model outlines complicated boundaries of different modal objects much more accurately than other known counterparts.
Computational Imaging and Vision, 2000
This TR discusses aspects of using ground control to validate the computational terrain reconstru... more This TR discusses aspects of using ground control to validate the computational terrain reconstruction. Image features provided for stereo matching allow to deduce simple confidence measures for reconstructed terrains, and only sufficiently confident terrain points should be validated by the available control data.
Lecture Notes in Computer Science, 1998
We address modelling of stochastic image textures by Gibbs random fields with a translation invar... more We address modelling of stochastic image textures by Gibbs random fields with a translation invariant structure of multiple pairwise pixel interactions. The characteristic interaction structure aPd strengths (Gibbs potentials) are learnt from a given training sample by analytic and stochastic approximation of the unconditional or conditional maximum likelihood estimates of the potentials. The interaction structure is revealed by a model-based interaction map showing the relative contributions of each interaction to a total Gibbs energy. Features of the inter:ration maps are discussed and illustrated by experiments with various natural textures.
Lecture Notes in Computer Science, 2001
Lecture Notes in Computer Science, 1998
Two parameter learning schemes for Gibbs random field image models with translation invariant mul... more Two parameter learning schemes for Gibbs random field image models with translation invariant multiple pairwise pixel interactions are discussed. The schemes allow to estimate both the interaction structure and strengths (Gibbs potentials) from a given training sample. The first scheme is based on the unconditional MLE of the potentials. The estimates are specified in an implicit form and can be obtained in three steps: (i) an analytic first approximation of the potentials for a big many possible neighbours, (ii) a search for most characteristic neighbours, and (iii) a stochastic approximation refinement of the estimates for a chosen set of neighbours. The second scheme uses the conditional MLE suggesting that the training sample has the least upper bound (top rank) in its total Gibbs energy within the parent population. This scheme allows to deduce an explicit, to scaling factors, analytic form of the potentials. Then only the scaling factors have to be learnt using their MLE in a like three-step manner. The conditional MLE of the potentials seems to be close to the unconditional ones and extends capabilities of the Gibbs image models.
Lecture Notes in Computer Science, 2001
A novel approach to computational binocular stereo based on the Neyman-Pearson criterion for disc... more A novel approach to computational binocular stereo based on the Neyman-Pearson criterion for discriminating between statistical hypotheses is proposed. An epipolar terrain profile is reconstructed by maximizing its likelihood ratio with respect to a purely random profile. A simple generative Markov-chain model of an image-driven profile that extends the model of a random profile is introduced. The extended model relates transition probabilities for binocularly and monocularly visible points along the profile to grey level differences between corresponding pixels in mutually adapted stereo images. This allows for regularizing the ill-posed stereo problem with respect to partial occlusions.
Lecture Notes in Computer Science, 2001
Supervised segmentation of piecewise-homogeneous image textures using a modified conditional Gibb... more Supervised segmentation of piecewise-homogeneous image textures using a modified conditional Gibbs model with multiple pairwise pixel interactions is considered. The modification takes into account that interregion interactions are usually different for the training sample and test images. Parameters of the model learned from a given training sample include a characteristic pixel neighbourhood specifying the interaction structure and Gibbs potentials giving quantitative strengths of the pixelwise and pairwise interactions. The segmentation is performed by approaching the maximum conditional likelihood of the desired region map provided that the training and test textures have similar conditional signal statistics for the chosen pixel neighbourhood. Experiments show that such approach is more efficient for regular textures described by different characteristic long-range interactions than for stochastic textures with overlapping close-range neighbourhoods.
Lecture Notes in Computer Science, 2002
Spatially homogeneous regular mosaics are image textures formed as a tiling, each tile replicatin... more Spatially homogeneous regular mosaics are image textures formed as a tiling, each tile replicating the same texel. Assuming that the tiles have no relative geometric distortions, the orientation and size of a rectangular texel can be estimated from a model-based interaction map (MBIM) derived from the Gibbs random field model of the texture. The MBIM specifies the structure of pairwise pixel interactions in a given training sample. The estimated texel allows us to quickly simulate a large-size prototype of the mosaic.
Advances in Pattern Recognition, 2000
Gibbs models with multiple pairwise pixel interactions permit us to estimate characteristic inter... more Gibbs models with multiple pairwise pixel interactions permit us to estimate characteristic interaction structures of spatially homogeneous image textures. Interactions with partial energies over a particular threshold form a basic structure that is sufficient to model a specific group of stochastic textures. Another group, referred here to as regular textures, permits us to reduce the basic structure in size, providing only a few primary interactions are responsible for this structure. If the primary interactions can be considered as statistically independent, a sequential learning scheme reduces the basic structure and complements it with a fine structure describing characteristic minor details of a texture. Whereas the regular textures are described more precisely by the basic and fine interaction structures, the sequential search may deteriorate the basic interaction structure of the stochastic textures.
Uploads
Papers by Georgy Gimelfarb