Abstract
In recent years, image processing and computer vision have played an important role in many scientific and technological areas mainly because modern society highlights vision over other senses. Throughout the time, application requirements and complexity have been increasing. Due to the fact that in many cases solutions depend on intrinsic characteristics of problems it is difficult to propose a universal model. In parallel, advances in understanding the human visual system have allowed to use sophisticated image models that incorporate simple phenomena, which occur in early stages of the visual system. Such phenomena suit visual stimuli for further processing. This thesis aims to investigate characteristics of vision such as over-representation and orientation of receptive fields in order to propose bio-inspired image models. Starting from studies of Gabor, Daugman, and Hubel we present an overcomplete image model that takes advantage of redundant information. Furthermore, we performed a comparison with several models from the state-of-art. Our proposal is based on Gabor filters and optimizes redundant information; such an information is distributed uniformly onto frequency bands and orientations with promising results. It is well known that Gabor models generated high-dimensional representations, therefore, we included a step where data dimension is reduced using Fisher theory and kernel methods. This step leads to a better characterization of visual scenes. In order to validate our method, we performed several experiments of segmentation and classification using synthetic textures. The last part of this dissertation claims that the combination of global and local descriptors will provide robust features that lead to an improvement in the classification rate. We included a study of local descriptors, specifically based on local binary patterns, and introduced a combined scheme for classifying textures of lung emphysema. The results of the experiments are consistent with the theory and demonstrate the effectiveness of our proposal.
Author
Download
Citation info
If you find useful this document, please include a citation in your work :0)
@PHDTHESIS{NAVA2013,
author = {Rodrigo Nava},
title = {Overcomplete Image Representations for Texture Analysis},
school = {Universidad Nacional Aut\’onoma de M\’exico},
year = {2013},
month = {10}
}
Rodrigo Nava 