Abstract
The paper overviews a range of novel techniques recently developed for efficient detection and further analysis of data in visual databases. A particular attention is paid to the issue that, until recently, has been considered too complex and ill-defined, i.e. identification of unspecified similar fragments in images of unpredictable and unknown contents. We discuss the following techniques that can be instrumental in solving this problem: (1) detection of image fragments related by affine transformations; (2) detection of image fragments related topological mappings; (3) accurate shape estimate of detected similar image fragments and (4) automatic classification of fragments into classes of objects of similar visual properties. The paper presents the theoretical background and the implementation results. Additionally, a demo of a real-time system for detection of fragments similar to database images in a video stream will be presented during the conference.