Motion vector refinement for MPEG-2 to H.264 video transcoding
A method for transcoding that includes (A) generating a decoded frame by decoding an MPEG-2 input video stream, the decoded frame including decoded macroblocks; (B) determining a search center for encoding a current macroblock corresponding to a pair of the decoded macroblocks on consecutive macroblock rows, when (i) the encoding uses a predictive field mode and (ii) a current field being encoded is a second field of a current frame that has a first field as a reference field, the search center is a temporally scaled version of a decoded motion vector from an upper or a lower macroblock in the pair; (C) generating a refined motion vector by searching in a temporal search direction about the search center; and (D) generating an H.264 output video stream by encoding the current macroblock based on the refined motion vector.
1. A method for transcoding, comprising the steps of:
(A) generating a decoded frame by decoding an input video stream in an MPEG-2 format, said decoded frame including a plurality of decoded macroblocks;
(B) determining a current search center to be used in encoding a current macroblock into an H.264 format, said current macroblock corresponding to a particular one of said decoded macroblocks, wherein when said particular macroblock has an intra-prediction, (i) said current search center comprises a previously determined search center that temporally matches a temporal search direction and (ii) said previously determined search center is spatially adjoining said particular macroblock;
(C) generating a refined motion vector by searching in said temporal search direction about said current search center; and
(D) generating an output video stream in said H.264 format by said encoding of said current macroblock based on said refined motion vector, wherein said encoding into said H.264 format utilizes macroblock adaptive field/frame encoding.
2. The method according to claim 1 , wherein in step (B) when said particular macroblock has (i) at least one motion vector matching said temporal search direction and (ii) and a frame motion compensation, said current search center comprises a frame motion vector of said particular macroblock.
3. The method according to claim 1 , wherein in step (B) when said particular macroblock has (i) at least one motion vector matching said temporal search direction and (ii) and a field motion compensation, said current search center comprises a vertically scaled version of a field motion vector of a top field of said particular macroblock.
4. The method according to claim 1 , wherein in step (B) when said particular macroblock has (i) no motion vectors matching said temporal search direction, (ii) at least one motion vector opposite said temporal search direction and (iii) a frame motion compensation, said current search center comprises a mirrored version of a frame motion vector of said particular macroblock.
5. The method according to claim 1 , wherein in step (B) when said particular macroblock has (i) no motion vectors matching said temporal search direction, (ii) at least one motion vector opposite said temporal search direction and (iii) a field motion compensation, said current search center comprises a scaled and mirrored version of a field motion vector of a top field of said particular macroblock.
6. A system comprising:
a decoder circuit configured to generate a decoded frame by decoding an input video stream in an MPEG-2 format, said decoded frame including a plurality of decoded macroblocks; and
an encoder circuit configured to (i) determine a current search center to be used in encoding a current macroblock into an H.264 format, said current macroblock corresponding to a particular one of said decoded macroblocks, wherein when said particular macroblock has an intra-prediction, (a) said current search center comprises a previously determined search center that temporally matches a temporal search direction and (b) said previously determined search center is spatially adjoining said particular macroblock, (ii) generate a refined motion vector by searching in said temporal search direction about said current search center and (iii) generate an output video stream in said H.264 format by said encoding of said current macroblock based on said refined motion vector, wherein said encoding into said H.264 format utilizes macroblock adaptive field/frame encoding.