Catálogo de publicaciones - libros

This paper proposes a video retrieval system from compressed outdoor video surveillance databases. The aim is to extract moving objects from frames provided by MPEG video stream in order to classify them into predefined categories according to image-based properties, and then robustly index them. The principal idea is to combine between useful properties of metrical classification and the notion of temporal consistency. Fuzzy geometry classification is used in order to provide an efficient method to classify motion regions into three generic categories: pedestrian, vehicle and no identified object. The temporal consistency provides a robust classification to changes of objects appearance and occlusion of object motion. The classified motion regions are used as templates for metrical training algorithms and as keys for tree indexing technique.

In this paper, a novel and efficient automatic image categorization system is proposed. This system integrates the MIL-based and global-feature-based SVMs for categorization. The IPs (Instance Prototypes) are derived from the segmented regions by applying MIL on the training images from different categories. The IPs-based image features are further used as inputs to a set of SVMs to find the optimum hyperplanes for categorizing training images. Similarly, global image features, including color histogram and edge histogram, are fed into another set of SVMs. For each test image, two sets of image features are constructed and sent to the two respective sets of SVMs. The decision values from two sets of SVMs are finally incorporated to obtain the final categorization results. The empirical results demonstrate that the proposed system outperforms the peer systems in terms of both efficiency and accuracy.

We present a modified version of the Generic Fourier Descriptor (GFD) that operates on edge information within natural images from the COREL image database for the purpose of shape-based image retrieval. By incorporating an edge-texture characterization (ETC) measure, we reduce the complexity inherent in oversensitive edge maps typical of most gradient-based detectors that otherwise tend to contaminate the shape feature description. We find that the proposed techniques not only improve overall retrieval in terms of shape, but more importantly, provide for a more accurate similarity ranking of retrieved results, demonstrating greater consideration for dominant internal and external shape details.

This paper demonstrates an approach to image retrieval by classifying images into different semantic categories and using probabilistic similarity measures. To reduce the semantic-gap based on low-level features, a relevance feedback mechanism is also added, which refines the query parameters to adjust the matching functions. First and second order statistical parameters (mean and covariance matrix) are pre-computed from the feature distributions of predefined categories on multivariate Gaussian assumption. Statistical similarity measure functions utilize these category specific parameters based on the online prediction of a multi-class support vector machine classifier. In relevance feedback, user selected positive or relevant images are used for calculating new query point and updating statistical parameters in each iteration. Whereas, most prominent relevant and non-relevant category specific information are utilized to modify the ranking of the final retrieved images. Experimental results on a generic image database with ground-truth or known categories are reported. Performances of several probabilistic distance measures are evaluated, which show the effectiveness of the proposed technique.

The aim of this work is to propose a method for recovering the 3D geometry of a video sequence taken from a pair of stereo cameras. The cameras are rigidly situated in a fixed position and there are some objects which are moving in front of them. Our method estimates the displacements of objects and the 3D structure of the scene. We establish a temporal constraint that relates the computation of the optical flow and the estimation of disparity maps. We use an energy minimisation approach that yields a system of partial differential equations (PDE) which is solved by means of a gradient descent technique.

Three-dimensional planar profile sampling of surfaces is a very common method of structural recovery in 3D scanning. In handheld 3D scanners, this has scarcely ever been taken into account resulting in poor precision ratings. Therefore, in this text we will describe a novel use of the profiling geometrical context to derive an intuitive and physically meaningful approach on solving the 3D profile registration problem. We will finish by describing the global optimisation algorithm and by showing experimental results achieved with a 3D scanner prototype comprising a camera, a laser-plane projector and a pose sensor.

This paper presents a method for estimating the orientation of planar text surfaces using the edge-direction distribution (EDD) extracted from the image as input to a neural network. We consider canonical rotations and we developed a mathematical model to analyze how the EDD changes with the rotation angle under orthographic projection. In order to improve performance and solve quadrant ambiguities, we adopt an active-vision approach by considering a pair of images (instead of only one) with a slight rotation difference between them. We then use the difference between the two EDDs as input to the network. Starting with camera-captured front-parallel images with text, we apply single-axis synthetic rotations to verify the validity of the EDD transform model and to train and test the network. The presented text-pose estimation method is intended to provide navigation guidance to a mobile robot capable of reading the textual content encountered in its environment.

We describe a new stereoscopic system based on a multispectral camera and an -Projector. The novel concept we want to show consists in the use of multispectral information for 3-scenes reconstruction. Each 3 point is linked to a curve representing the spectral reflectance. This latter is a physical representation of the matter and presents the advantage over color information, which is perceptual, that it is independent from both illuminant and observer. We first present an easy methodology to geometrically and spectrally calibrate such a system. We then describe an algorithm for recovering 3 coordinates based on triangulation and an algorithm for reflectance curves reconstruction based on neural networks. The results are encouraging. they confirm the feasibility of such a system and in the same time enable some applications like simulating illumination change.

This paper describes a real-time stereo depth measurement image processing system. This system uses Xilinx Virtex-II Series XC2V3000 FPGA and generates 8-bit sub-pixel disparities on 640 by 480 resolution images at video rate (60 frames/sec) with maximum disparity ranges of up to 128 pixels. The implemented stereo matching algorithm finds a minimum of window-based sum of absolute difference (SAD) operation. And the preprocessing, scale transformation and final stage compensation technique are adopted for maximizing the wide disparity range detection. The proposed vision system is suitable for real-time range estimation and robot navigation applications.

We present an application of a family of affine diagrams to the detection of three-dimensional sampled structures embedded in a perturbated background. This family of diagrams is an extension of the Voronoi diagram, namely the anisotropic diagrams. These diagrams are defined by using a parameterized distance whose unit ball is an ellipsoidal one. The parameters, upon which depends this distance, control the elongation and the orientation of the associated ellipsoidal ball. Based on these diagrams, we define the three-dimensional anisotropic -shape concept. This concept is an extension of the Euclidean one, it allows us to detect structures, as straight lines and planes, in a given direction. The detection of a more general polyhedral structure is obtained by merging several anisotropic -shapes, computed for different orientations.