Motion estimation with motion vector penalty
AssigneeTexas Instruments Incorporated
InventorsAkira Osamoto, Toshio Takamizawa
Life of this patent
Estimating the term…
Abstract
The MPEG motion estimation process is improved by the introduction of the motion vector penalty. The motion vector employed to encode a macroblock takes into consideration the number of bits needed to encode the macroblock with the selected motion vector. This consideration includes a sum of the residual error and the cost to encode the candidate motion vector. This provides an optimization of the bit allocation and vector type selection. This optimization results in a significantly improved picture quality.
Claims4
A method of motion estimation in a digital video compression system comprising the steps of:
- segmenting the target picture into a grid of macroblocks;
- for each macroblock
- choosing for test a plurality of candidate motion vectors according to an implemented search criteria,
- for each candidate motion vector calculating a residual error between the macroblock and a reference macroblock displaced by a current candidate motion vector,
- for each candidate motion vector estimating a cost of coding said current candidate motion vector dependent upon a difference between said candidate motion vector and a motion vector from an immediately previous macroblock including separately determined cost of coding a horizontal difference of motion vectors and cost of coding a vertical difference of motion vectors,
- selecting one candidate motion vector providing a least number of bits to encode the macroblock dependent upon said calculated residual error and said cost of coding, and
- encoding the macroblock using the one selected candidate motion vector.
A digital video apparatus comprising:
- a source of a digital video signal;
- a digital signal processor connected to said source of a digital video signal programmed to:
- choose for test a plurality of candidate motion vectors according to an implemented search criteria,
- for each candidate motion vector calculate a residual error between the macroblock and a reference macroblock displaced by a current candidate motion vector,
- for each candidate motion vector estimate a cost of coding said current candidate motion vector dependent upon a difference between said candidate motion vector and a motion vector from an immediately previous macroblock including separately determined cost of coding a horizontal difference of motion vectors and cost of coding a vertical difference of motion vectors,
- select one candidate motion vector providing a least number of bits to encode the macroblock, and
- encode the macroblock using the one selected candidate motion vector.
Dependencies are inferred from the claim text.