Semantic feature extraction using Mpeg macro-block classification

In this paper, we present some first results in the extraction of semantic features from video sequences. Our approach is based on the classification of Mpeg DCT macro-blocks. Although it is clear that using macro-blocks imposes severe restrictions on the analysis accuracy of the image, it has the advantage of avoiding the complete decoding of
the Mpeg stream. Our objective is to evaluate the quality of the Semantic Feature Extraction that can be obtained with this direct approach, to serve as a comparative baseline with more elaborate approaches.