Fast transcoding architectures for insertion of non-regular shaped objects in the compressed DCT-domain

摘要

This paper addresses the problem concerned with the insertion of non-regular shaped objects, such as logos or subtitles, in compressed video signals. To achieve this objective, a different approach from the usual pixel-domain compositing operation is adopted, in order to avoid the presence of undesired semi-transparent rectangular regions around the inserted objects. The transposition of this technique to the compressed DCT-domain is presented, as well as the integration of the logo insertion module in several architectures of compressed domain video transcoders. A detailed study of the main characteristics of the proposed DCT-domain transcoders, with a special emphasis on the required computational load and on the obtained coding quality (PSNR) of the processed video sequences shows that DCT-domain insertion architectures can offer significant advantages, both in terms of subjective quality and computational cost. Moreover, the distinctive features presented by the proposed architectures provide the means to adapt the insertion scheme to the particular requirements of the target application.