ROI extraction of MPEG-2 video using partial decoding for MPEG-21 digital item adaptation

Abstract

Most of multimedia transcoding techniques, especially video transcoding, focus on spatial reduction by down-sampling according to the physical size of a display, temporal reduction of video frame frequency by frame dropping according to network condition and visual quality reduction by dropping signal-to-noise ratio with coarse quantization on DCT coefficients. User``s interest in contents has not been considered into account when being transcoded. In this thesis, we present a semantic transcoding method that extracts a region of interest(ROI) in image or video contents. The ROI extraction is done by partial decoding in MPEG-2 video bitstream domain. We also incorporate this ROI extraction scheme into an adaptation engine in MPEG-21 Digital Item Adaptation framework. For the ROI extraction in MPEG-2 video bitstreams, we design a ROI extraction scheme that extracts an ROI for any given position in a video frame in the bitstream domain. By using a partial decoding in the ROI extraction, we avoid large computational overhead by full decoding and re-encoding for transmission and storage. Also, by performing ROI extraction in the bitstream domain, degradation of visual quality does not occur otherwise it occurs for full decoding and re-encoding. The experiment result shows that the proposed ROI extraction scheme with partial decoding reduces computational time taken for ROI extraction by 34 percents by preserving the original encoded visual quality of the MPEG-2 encoded video.