Catálogo de publicaciones - libros

This work reports a new methodology to develop a tumor grade diagnostic system, which is based on the integration of experts’ knowledge with image analysis techniques. The proposed system functions in two-levels and classify tumors according to their histological grade in three categories. In the lower-level, values of certain histopathological variables are automatically extracted by image analysis methods and feed the related concepts of a Fuzzy Cognitive Map (FCM) model. FCM model on the upper level interacts through a learning procedure to calculate the grade scores. Final class accuracy is estimated using the k-nearest classifier. The integrated FCM model yielded an accuracy of 63.63%, 72.41% and 84.21% for tumors of grade I, II, and III respectively. Results are promising, revealing new means for mining quantitative information and encoding significant concepts in decision process. The latter is very important in the field of computer aided diagnosis where the demand for reasoning and understanding is of main priority.

Finding the distance of an object in a scene from intensity images is an essential problem in many applications. In this work, we present a novel method for depth recovery from a single motion and defocus blurred image. Under the assumption of uniform linear motion between the camera and the scene during finite exposure time, both the pinhole model and the camera with a finite aperture are considered. The blur extent is estimated by intensity profile analysis and focus measurement of the deblurred images. The proposed method has been verified experimentally using edge images.

We present an experimental analysis of four different ways of constructing three-dimensional statistical models of faces using Cartesian coordinates, namely: height, surface gradient, azimuthal angle and one based on Fourier domain basis functions. We test the ability of each of the models for dealing with information provided by shape-from-shading. Experiments show that representations based on directional information are more robust to noise than representations based on height information. Moreover, the method can be operated using a simple non-exhaustive parameter adjustment procedure and ensures that the recovered surface satisfies the image irradiance equation as a hard constraint subject to integrability conditions.

A new method for the catadioptric omnidirectional stereo vision with single camera is presented in this paper. The proposed method uses a concave lens with a convex mirror. Since the optical part of the proposed method is simple and commercially available, the resulting omnidirectional stereo system is compact and cost-effective. The closed-form solution for 3D distance is derived based on the simple optics including the reflection and the refraction of the convex mirror and the concave lens. The compactness of the system and the simplicity of the image processing make the omnidirectional stereo system appropriate for real-time applications such as autonomous navigation of a mobile robot or the object manipulation. In order to verify the feasibility of the proposed method, an experimental prototype is implemented.

This paper presents a neural network (NN) to recover three-dimensional (3D) shape of an object from its multiple view images. The object may contain non-overlapping transparent and opaque surfaces. The challenge is to simultaneously reconstruct the transparent and opaque surfaces given only a limited number of views. By minimizing the pixel error between the output images of this NN and teacher images, we want to refine vertices position of an initial 3D polyhedron model to approximate the true shape of the object. For that purpose, we incorporate a ray tracing formulation into our NN’s mapping and learning. At the implementation stage, we develop a practical regularization learning method using texture mapping instead of ray tracing. By choosing an appropriate regularization parameter and optimizing using hierarchical learning and annealing strategies, our NN gives more approximate shape.

We develop a new model for the scattering of light from layered rough surfaces. The model contains a surface scattering component together with a subsurface scattering component. The former component corresponds to the roughness on the upper surface boundary and is modeled using the modified Beckmann model. The latter component accounts for both refraction due to Fresnel transmission through the layer and rough (Beckmann) scattering at the lower layer boundary. Depending on the type of surface, the contributions of the two scattering components to the total outgoing radiance can vary dramatically for different materials. This behavior is conveniently captured by allowing a balance parameter in the model. Our experiments with BRDF data from several types of surface and with different scales of roughness confirm that the new model outperforms alternative variants of the Beckmann model together with several alternative reflectance models.

We propose models based on Gabor functions to address two related aspects in the object recognition problem: interest point selection and classification. We formulate the interest point selection problem by a cascade of bottom-up and top-down stages. We define a novel type of top-down saliency operator to incorporate low-level object related knowledge very soon in the recognition process, thus reducing the number of canditates. For the classification process, we represent each interest point by a vector of Gabor responses whose parameters are automatically selected. Both the selection and classification procedures are designed to be invariant to rotations and scaling. We apply the approach to the problem of facial landmark classification and present experimental result illustrating the performance of the proposed techniques.

In this paper, we present an extended scheme of human action recognition with nearness information between hands and other body parts for the purpose of automatically analyzing nonverbal actions of human beings. First, based on the principle that a human action can be defined as a combination of multiple articulation movements, we apply the inference of stochastic grammars. We measure and quantize each human action in 3D coordinates and make two sets of 4-chain-code for xy and zy projection planes, so that they are appropriate for the stochastic grammar inference method. Next, we extend the stochastic grammar inferring method by applying nearness information. We confirm that various physical actions are correctly classified against a set of real-world 3D temporal data with this method in experiments. Our experiments show that this extended method reveals comparatively successful achievement with a 92.7% recognition rate of 60 movements of the upper body.

This paper presents a robust technique for a real time estimation of both camera’s position and orientation—referred as pose. A commercial stereo vision system is used. Unlike previous approaches, it can be used either for urban or highway scenarios. The proposed technique consists of two stages. Initially, a compact 2D representation of the original 3D data points is computed. Then, a RANSAC based least squares approach is used for fitting a plane to the road. At the same time, relative camera’s position and orientation are computed. The proposed technique is intended to be used on a driving assistance scheme for applications such as obstacle or pedestrian detection. Experimental results on urban environments with different road geometries are presented.

With the rapid increase of available 3D models, content-based 3D retrieval is attracting more and more research interests. One of the key problems in content-based 3D retrieval is to extract discriminative features for measuring the similarity and dissimilarity between different shapes. In this paper, we define 3D Krawtchouk moments for 3D shape analysis and retrieval. Differing with 3D Zernike moments deduced from continuous orthogonal polynomials, the basis of 3D Krawtchouk moments is discrete orthogonal polynomial. It has some interesting property for describing shape information and retrieving 3D models, such as multi-resolution, high-computation, simplification and so on. To verify the advantage of 3D Krawtchouk moments, experiments are carried out to compare the retrieving performance based on Krawtchouk moments and Zernike moments. The results have proven that Krawtchouk moments can achieve better retrieving accuracy and efficiency.