Content Based Video Retrival System for Mexican Culture Heritage Based on Object Matching and Local-Global Descriptors

Multimedia data and networking technologies have had a highly growing during the last decade, with these changes users have changed from text to content based video retrieval systems due to its better performance. We propose a fast content-based video retrieval system which involves the combination of a local descriptor obtained from the speeded-up robust feature algorithm together with an effective and fast object matching operation. To save computational time, compressed video data are partially decoded in order to get discrete cosine transform coefficients of key frames, which are used to obtain sub-block coefficients and a down-sampling version of frames. The preliminary results are ranking using an efficient color descriptor based on color correlogram and dominant color descriptors. To measure the performance of the proposed technique the precision and recall metrics are used. The experimental results show the accuracy of the proposed method applied to a database of Mexican Culture Heritage videos.