Electronic Letters on Computer Vision and Image Analysis (ELCVIA - Universitat Autònoma de Barcelona)
Not a member yet
343 research outputs found
Sort by
Matching Local Invariant Features with Contextual Information: An Experimental Evaluation
Get PDFThe main advantage of using local invariant features is their local character which yields robustness to occlusion and varying background. Therefore, local features have proved to be a powerful tool for finding correspondences between images, and have been employed in many applications. However, the local character limits the descriptive capability of features descriptors, and local features fail to resolve ambiguities that can occur when an image shows multiple similar regions. Considering some global information will clearly help to achieve better performances. The question is which information to use and how to use it. Context can be used to enrich the description of the features, or used in the matching step to filter out mismatches. In this paper, we compare different recent methods which use context for matching and show that better results are obtained if contextual information is used during the matching process. We evaluate the methods in two applications: wide baseline matching and object recognition, and it appears that a relaxation based approach gives the best results
Principal Deformations Modes of Articulated Models for the Analysis of 3D Spine Deformities
Get PDFArticulated models are commonly used for recognition tasks in robotics and in gait analysis, but canalso be extremely useful to develop analytical methods targeting spinal deformities studies. The threedimensionalanalysis of these deformities is critical since they are complex and not restricted to a givenplane. Thus, they cannot be assessed as a two-dimensional phenomenon. However, analyzing large databasesof 3D spine models is a difficult and time-consuming task. In this context, a method that automatically extractsthe most important deformation modes from sets of articulated spine models is proposed.The spine was modeled with two levels of details. In the first level, the global shape of the spine wasexpressed using a set of rigid transformations that superpose local coordinates systems of neighboring vertebrae.In the second level, anatomical landmarks measured with respect to a vertebra’s local coordinatesystem were used to quantify vertebra shape. These articulated spine models do not naturally belong to avector space because of the vertebral rotations. The Fréchet mean, which is a generalization of the conventionalmean to Riemannian manifolds, was thus used to compute the mean spine shape. Moreover, ageneralized covariance computed in the tangent space of the Fréchet mean was used to construct a statisticalshape model of the scoliotic spine. The principal deformation modes were then extracted by performing aprincipal component analysis (PCA) on the generalized covariance matrix.The principal deformations modes were computed for a large database of untreated scoliotic patients.The obtained results indicate that combining rotation, translation and local vertebra shape into a unifiedframework leads to an effective and meaningful analysis method for articulated anatomical structures. Thecomputed deformation modes also revealed clinically relevant information. For instance, the first mode ofdeformation is associated with patients’ growth, the second is a double thoraco-lumbar curve and the thirdis a thoracic curve. Other experiments were performed on patients classified by orthopedists with respect toa widely used two-dimensional surgical planning system (the Lenke classification) and patterns relevant tothe definition of a new three-dimensional classification were identified. Finally, relationships between localvertebrae shapes and global spine shape (such as vertebra wedging) were demonstrated using a sample of3D spine reconstructions with 14 anatomical landmarks per vertebra.KeyWords: Shape Analysis, Articulated Models, Spinal Deformities, Scoliosis, 3D Reconstruction, Surgical Classifications
Facial Emotional Classifier For Natural Interaction
Get PDFThe recognition of emotional information is a key step toward giving computers the ability to interact more naturally and intelligently with people. We present a simple and computationally feasible method to perform automatic emotional classification of facial expressions. We propose the use of a set of characteristic facial points (that are part of the MPEG4 feature points) to extract relevant emotional information (basically five distances, presence of wrinkles in the eyebrow and mouth shape). The method defines and detects the six basic emotions (plus the neutral one) in terms of this information and has been fine-tuned with a database of more than 1500 images. The system has been integrated in a 3D engine for managing virtual characters, allowing the exploration of new forms of natural interaction
Development of a machine vision system for a real time precision sprayer
Get PDFIn the context of precision agriculture, we have developed a machine vision system for a real timeprecision sprayer. From a monochrome CCD camera located in front of the tractor, the discriminationbetween crop and weeds is obtained with an image processing based on spatial information using a Gaborfilter. This method allows to detect the periodic signals from the non periodic one and it enables to enhancethe crop rows whereas weeds have patchy distribution. Thus, weed patches were clearly identified by ablob-coloring method. Finally, we use a pinhole model to transform the weed patch coordinates image inworld coordinates in order to activate the right electro-pneumatic valve of the sprayer at the right moment
GPU-Based Optimization of a Free-Viewpoint Video System
Get PDFWe present a method for optimizing the reconstruction and rendering of 3D objects from multiple images by utilizing the latest features of consumer-level graphics hardware based on shader model 4.0. We accelerate visual hull reconstruction by rewriting a shape-from-silhouette algorithm to execute on the GPU\u27s parallel architecture. Rendering is optimized through the application of geometry shaders to generate billboarding micr facets textured with captured images. We also present a method for handling occlusion in the camera selection process that is optimized for execution on the GPU. Execution time is further improved by rendering intermediate results directly to texture to minimize the number of data transfers between graphics and main memory. We show our GPU based system to be significantly more efficient than a purely CPUbased approach, due to the parallel nature of the GPU, while maintaining graphical quality
Enhancing Sensor Measurements throughWide Baseline Stereo Images
Get PDFIn this paper, we suggest an algorithm to enhance the accuracy of sensor measurements representingcamera parameters. The process proposed is based solely on a pair of wide baseline (or sparse view) images.We use the so-called JUDOCA operator to extract junctions. This operator produces junctions in termsof locations as well as orientations. Such an information is used to estimate an affine transformation matrix,which is used to guide a variance normalized correlation process that produces a set of possible matches.The fundamental matrix can be easily estimated using the so-called RANSAC scheme. Consequently, theessential matrix can be derived given the available calibration matrix. The essential matrix is then decomposedusing Singular Value Decomposition. In addition to a translation vector, this decomposition results ina rotation matrix with accurate rotation angles involved. Mathematical derivation is done to extract anglesfrom the rotation matrix and express them in terms of different rotation systems
Class Specific Object Recognition using Kernel Gibbs Distributions
Get PDFFeature selection is crucial for effective object recognition. The subject has been vastly investigated in the literature, with approaches spanning from heuristic choices to statistical methods, to integration of multiple cues. For all these techniques the final result is a common feature representation for all the considered object classes. In this paper we take a completely different approach, using class specific features. Our method consists of a probabilistic classifier that allows us to use separate feature vectors, selected specifically for each class. We obtain this result by extending previous work on Class Specific Classifiers and Kernel Gibbs distributions. The resulting method, that we call Kernel-Class Specific Classifier, allows us to use a different kernel for each object class by learning it. We present experiments of increasing level of difficulty, showing the power of our approach
List of Reviewers
No full textThe following is a list of reviewers who from January 2008 to December 2008, have contributed their time and talent to the success of the publication of the ELECTRONIC LETTERS ON COMPUTER VISION AND IMAGE ANALYSIS
Multiresolution Shape Codes with Applications to Image Retrieval
Get PDFA novel technique on shape coding of an object in a binary digital image, based on the tight isotheticpolygonal covers of the object in a multiresolution background, is proposed. This technique would be useful in various analyses and applications related with digital images. To demonstrate the power and usefulness of such shape codes, we have also proposed an image retrieval scheme based on shape codes. The elegance of the scheme on shape codes lies in capturing the shape of the object(s) present in an image from its gross appearance to its finer details by a set of isothetic polygons, in a hierarchical manner. The inherent tolerance present in such a cast of isothetic polygonal shape enables the shape codes of two objects resemble closely as the grid resolution becomes finer and finer, provided the objects are of similar shapes. The method is very fast because it does not involve any complex computations, and requires only integer-based comparisons and additions. Experimental results demonstrate the strength and efficiency of the proposed scheme
Gray-level Texture Characterization Based on a New Adaptive Nonlinear Auto-Regressive Filter
Get PDFIn this paper, we propose a new nonlinear exponential adaptive two-dimensional (2-D) filter for texturecharacterization. The filter adaptive coefficients are updated with the Least Mean Square (LMS) algorithm. Theproposed nonlinear model is used for texture characterization with a 2-D Auto-Regressive (AR) adaptive model. Themain advantage of the new nonlinear exponential adaptive 2-D filter is the reduced number of coefficients used tocharacterize the nonlinear image regarding the 2-D second-order Volterra model. Whatever the degree of the nonlinearity,the problem results in the same number of coefficients as in the linear case. The characterization efficiency ofthe proposed exponential model is compared to the one provided by both 2-D linear and Volterra filters and the cooccurrencematrix method. The comparison is based on two criteria usually used to evaluate the features discriminatingability and the class quantification in characterization techniques. The first criterion is proposed to quantify theclassification accuracy based on a weighted Euclidean distance classifier. The second criterion is the characterizationdegree based on the ratio of ";;;;;;;between-class";;;;;;; variances with respect to ";;;;;;;within-class";;;;;;; variances of the estimatedcoefficients. Extensive experiments proved that the exponential model coefficients give better results in texturediscrimination than several other parametric characterization methods even in a noisy context.Key words: Image Analysis, 2-D nonlinear filter, 2-D adaptive filter, texture characterization