Semantic Video Adaptation using a Preprocessing Method for Mobile Environment

Video adaptation is critical for video encoding and streaming in mobile environment. Semantic video adaptation takes into account the degree of relevance of different part of the image and tries to adapt video content to meet wireless channel constraint. Making assumption that only changing region of the video is important to mobile user, we propose a real-time semantic level adaptation technique based on video preprocessing. Firstly moving foreground of a video frame is segmented from background scene. Then a low-pass filter is applied to the background of the video frame before it is encoded. By this way, background is blurred and video encoder is forced to allocate more bit rate to foreground part of the frame than the background thus improves the quality of semantically significant portion of the image. To improve robustness a user feedback strategy is employed enabling user to choose the degree of background blurring according their preferences. The simulation result shows that the proposed method improves the quality of encoded video by both subjective and objective measurement.