Author: Ángeles LÓPEZ
Supervisor: Filiberto PLA

Abstract

Stereoscopic vision includes a wide range of techniques for correspondence and three-dimensional reconstruction. Region-based techniques have been hardly addressed in the literature compared with edge-based techniques or area-based techniques. This is due mainly to the fact that the results of segmentation methods vary significantly from one image to another, which makes the matching of regions more difficult.

In this thesis, the problems generated by the differences in segmentation from both images of the stereo pair are faced by means of two different techniques. First, we use a classic graph-based technique, widely used in the literature, where we add a new preprocess stage, which is a graph-based step, in order to solve the segmentation problems. Second, we present a novel matching technique where only one of the images is segmented, and correspondences are searched in the other image by means of the minimization of an energy function and the application of some constraints on the depth of the regions. In this technique, correspondences, three-dimensional reconstruction and occlusions are obtained in an only cooperative process.

Keywords: stereoscopic vision, segmentation, correspondence, 3D reconstruction, occlusion detection, correlation, association graph, maximal clique, minimization of an energy function.