METHOD AND APPARATUS FOR VIDEO INTERMODAL TRANSCODING

摘要

A video intermodal transcoder converts a compressed bitstream formulated according a type-1 compression scheme to a type-2 compressed bitstream formulated according to a type-2 compression scheme. The transcoder includes an augmented type-1 decoder, a transcoder kernel, and an augmented type-2 encoder. The transcoder kernel performs processes of creating motion-vector candidates and pre-computing prediction errors for each cell of a predefined image coding block and for each candidate motion vector for repetitive use in evaluating various image partitions. In an implementation where the type-1 compression scheme follows the H.264 standard and the type-2 compression scheme follows the HEVC standard, the transcoder exploits the flexibility of the coding-tree structure and other HEVC features to significantly reduce the bit rate of the compressed bit stream. The pre-computation of prediction errors significantly reduces the processing effort, hence increases the throughput of the transcoder.

主权项

1. A method, implemented by a hardware processor accessing a memory device storing software instructions, for transcoding a succession of type-1 images compressed according to a first compression mode to type-2 images compressed according to a second compression mode, comprising:
extracting descriptors of a current type-1 image of said succession of type-1 images; generating a reconstructed bitstream of said current type-1 image; and for each predefined image region:
extracting motion-vector candidates of a current type-2 image from:
said current type-1 image;a reference type-2 image corresponding to a prior type-1 compressed image; anda synthesized part of said current type-2 image;creating a cache of prediction errors with reference to said reconstructed bitstream for each motion-vector candidate for each predefined cell of said each predefined image region;for a predefined segment of said each predefined image region and for each said motion-vector candidate determining, referencing said cache, a respective prediction error and selecting a preferred motion-vector candidate of least prediction error.