Motion vector size-compensation based method for very low bit-rate video coding

In this paper, a new method to achieve better compression efficiency in low bit-rate video coding is proposed. It is based on a global bit-rate reduction at a macroblock level, optimizing the number of bits to code each macroblock as a whole by means of motion vector and headers size compensation. The selection of the best motion vector and different coding modes for each block of the current picture is made depending not only on trying to choose the best prediction for the block, but also on the number of bits to code the associate headers, introducing some kind of penalization in the cost function. This method improves efficiency in video compression for all qualities, but especially for low-quality video coding, whose efficiency improvement can reach 17%. Its implementation is simple, and compatible with mast video-compression standards (H.263, MPEG, etc.). Results of the algorithm in a state-of-the-art H.263+ codec are presented, and demonstrate that the efficiency enhancement is achieved with minimal time-processing increase, and even decrease, in some conditions.