Motion vector estimation, motion picture encoding and storage

摘要

In the encoding of a frame (P3) in a digitised motion picture sequence, the frame is divided into blocks (A) of pixels and a motion vector (VA) found to predicts the contents of each block A from a block (A') in a previous frame (IO). Rather than evaluate many thousands of candidate vectors to find the best motion vector (VA), relatively few candidate vectors are evaluated, based on starting vector which is the estimated motion vector for a neighbouring block (LB(A)) in the same frame. To allow parallel processing of several pixel blocks, all the blocks of one row in the current frame are assigned to one processor (DSP), and a starting vector is obtained from a block (LB(A)) previously processed by the same processor. Further starting vectors are obtained if desired from neighbouring blocks (UR(B),DR(B)) in neighbouring frames of the sequence. The candidate vectors based on a starting vector may be very few in number, or may fill an area a few percent of the full search area. The method may favour candidate vectors which are more economical to encode.