Contribution to Image Segmentation and Integral Image Coding

Unsupervised image segmentation techniques based on the image textures are studied and developed in this thesis. First, the wavelet filters are considered as the mean for feature extraction and feature vector creation. The creation of the filters is studied as well as the filtered images segmented spectrum. Next, the analysis with directional filters for feature extraction is used. Emphasis is given on the directional filter construction and on the different way of spectrum segmentation for the analyzed images that is achieved. In the next step, after the feature generation, possible ways of feature reduction are considered, ways that do not affect the image separability as it is described by a properly selected separability criterion. Next, known clustering methods and the way they are incorporated in the unsupervised segmentation process are studied. Novel applications of those techniques that point out the contribution of this thesis are presented, and finally, a traversal scheme for the efficient integral image coding is suggested.