Department of Computer Engineering
S E M I N A R
Adaptation of Co-occurrence Matrices to the Low and Higher Level Image Representations
Mustafa Ilker Sarac
Tremendous amount of images on the web challenges the researchers for effective and efficient machine understanding of those in terms of retrieval and classification tasks. In the most traditional way image retrieval research has been done using textual tags provided by the images. However most of the time these textual tags are limited in quantities and does not capable of expressing the images in a way that human brain could perceive. That brings the need for the systems that are capable of interpret these images. At this point it is essential to engineer good low level image features(e.g. SIFT) to be used by these systems in order to fill the gap between human perception and the machine understanding of the images. In the literature many higher level representations are also proposed based on object or part based detectors.In this proposed thesis work, we aim to create a novel image descriptor by revisiting the idea of using co-occurrences for different levels of image representations in a texton-like environment.
DATE: 01 December, 2014, Monday @ 15:40