<title>Moving-object detection from MPEG coded data</title>

Text information extraction in images and video: a survey

“…Antani et al [28] and Gargi et al [32] utilize motion vectors in an MPEG-1 bit stream in the compressed domain for text tracking, which is based on the methods of Nakajama et al [62] and Pilu [63]. This method is implemented on the P and I frames in MPEG-1 video streams.…”

Face Tracking in the Compressed Domain

“…In the algorithms proposed in [6][7][8][9], objects can only be tracked to P-and I-frames, but not to B-frames.…”

Approximating optical flow within the MPEG-2 compressed domain

“…The rules used here to obtain the smooth motion field, unlike previous examples [6], [9], do not rely on future images since this work is targeted for real-time applications and any delay might be critical. Also, they are not restricted to any GOP format making it an adaptable tool.…”

“…Let be the 8 8 image block and be the block's DCT. The image block can be written in terms of using the standard inverse DCT (9) where , if and , otherwise (similarly for ). Note this reconstruction can be used for continuous values of and , not just integers and, therefore, the component of the spatial image gradient, , can be obtained by differentiating with respect to to get (10) and similarly for .…”

“…Very simple motion segmentation is achieved in [8] after median filtering the vector field. A better example is Nakajima's paper [9] where texture from I-frames is analyzed as a measure of confidence. Segmentation is also achieved in [10] after several stages of filtering applied to the vector field.…”