AN EFFICIENT OVERLAPPING BLOCK SELECTION METHOD IN MULTI-FRAME SKIPPING TRANSCODING

Abstract

This paper presents an overlapping block selection method to determine the dominant macroblock (MB) in multi-frame skipping transcoding to improve picture quality. A non-intra MB in a non-skipped frame should be predicted by a block in another non-skipped frame. The re-constructed motion vector (MV) between two non-skipped frames accumulates prediction error. As the number of frames between two non-skipped frames is increased, such prediction error, or called mismatch error, is also increased. To reduce the impact on the transcoding quality caused by the prediction error, we improve the traditional overlapping block selection method, and propose a most approximate overlapping block to help to determine a dominant MB. In the proposed method, the location of an overlapping block is calculated by adding the accumulated MVs and the position of the current MB. The accumulated MVs are not the real MVs for MBs but the approximate vectors for 16 × 16 overlapping blocks. Simulation shows that the proposed method has a higher performance than the traditional method.