Low complexity H.263 to H.264 video transcoding using motion vector decomposition

In this paper, we propose a simple architecture to form a low-complexity and high quality video transcoder to speed up the motion re-estimation for H.263 to H.264 video transcoding. Using the approximation of the prediction error from the incoming bitstream, the proposed transcoder is able to detect whether the performance of the incoming motion vector can be improved during the motion re-estimation process. By considering the diversity of the new set of the estimated motion vectors, the proposed video transcoder can control the decomposition of the motion vector dynamically, which can achieve low computational complexity. Besides, the multiple reference frames estimation is applied to a region which contain higher prediction error to further reduce the computational complexity during the multiple reference frames estimation. Results of our experimental work show that the proposed architecture produces similar picture quality with low computational complexity as compared with the conventional video transcoder.