Abstract |
During the past few years, there has been an explosion of the amount of the digitally stored visual information, due to the advances in the fields of communication technology and computer science. This fact intensifies the need for efficient coding, retrieval and transmission of this information. New video coding standards such as MPEG4 and MPEG7 make progress towards this direction. One key step in these standards is the segmentation of the images that consist the video, that is the partition of the image into disjoint sets of pixels that, hopefully, each corresponds to an object in the depicted scene. In this thesis we studied a recent class of segmentation algorithms based on energy minimization. The main idea is to construct a weighted graph where each vertex corresponds to a pixel in the image, and the edges are defined between neighboring pixels. The weights of each edge are carefully chosen so that, a minimum cut on the graph provides the minimum of the energy function and thus the partition we seek. We examined whether it is possible to extend these algorithms, in order to construct the graph with vertices that correspond to relatively small image segments that come from a preprocessing step. This reduces the running time and results in a more robust algorithm. We applied the extended algorithm to image sequences based on data from change detection, seeking a partition into two classes: static and mobile segments and the results are promising.
|