Indexing Techniques for Image and Video Databases: an approach based on Animate Vision Paradigm

In this dissertation some novel indexing techniques for video and image database based on “Animate Vision” Paradigm are presented and discussed. From one hand, it will be shown how, by embedding within image inspection algorithms active mechanisms of biological vision such as saccadic eye movements and fixations, a more effective query processing in image database can be achieved. In particular, it will be discussed the way to generate two fixation sequences from a query image I_q and a test image I_t of the data set, respectively, and how to compare the two sequences in order to compute a possible similarity (consistency) measure between the two images. Meanwhile, it will be shown how the approach can be used with classical clustering techniques to discover and represent the hidden semantic associations among images, in terms of categories, which, in turn, allow an automatic pre-classification (indexing), and can be used to drive and improve the query processing. Eventually, preliminary results will be presented and the proposed approach compared with the most recent techniques for image retrieval described in the literature. From the other one, it will be discussed how by taking advantage of such foveated representation of an image, it is possible to partitioning of a video into shots. More precisely, the shot-change detection method will be based on the computation, at each time instant, of the consistency measure of the fixation sequences generated by an ideal observer looking at the video. The proposed scheme aims at detecting both abrupt and gradual transitions between shots using a single technique, rather than a set of dedicated methods. Results on videos of various content types are reported and validate the proposed approach.