Research Seminar

Content based Video Retrieval (CBVR) based on salient features of shape and Trajectory information of a video object

Abstract

Retrieving video shots based on content has been a challenging task for many applications in the fields of digital video analysis and multimedia. Appropriate features derived from shape and motion cues of a video object are necessary to represent a video shot. An efficient matching criterion is also necessary to retrieve similar video shots from a database, with high accuracy.

In the first part of the talk, a method of representing salient points in a trajectory based of Gabor features will be described. This is a temporal-frequency representation of the trajectory, which involves the process of detecting a set of salient points from the peaks (locally) of the Gabor filter responses. Matching is performed based on edit distance and is shown to perform well for partial trajectory matching. We show how this method if efficient than other methods published in literature, using experiments on simulated dataset.

Then a method will be described to combine features from (i) shape contours (CSS based) and (ii) trajectory (Polynomial representation), to retrieve similar video shots from a real-world database. The shape representation is invariant to translation, rotation and scaling and robust with respect to noise. Trajectory matching incorporates visual distance, velocity dissimilarity and size dissimilarity for retrieval. The matching cost obtained by combining these two features will provide superior performance. The talk concludes with discussions on the challenges involved in future scope of work in this field.