| 摘要 |
Motion-compensated interframe-prediction can be realized by affine-transformation of suitably divided areas of an object image with a reduced amount of calculations and an improved coding efficiency. A motion-compensated interframe-prediction method obtains a predictive video-frame from motion-vectors of an input video-frame and a reference video-frame by translating blocks of processable area, determines a prediction-error estimation value of the predictive frame from an input video-frame, compares the determined prediction-error estimation value with a preset threshold value, selects a predictive video-frame obtained by translation (parallel displacement of coordinates) and encodes the prediction-error information when the prediction-error estimation value is smaller than the threshold, or, with the prediction-error estimation value exceeding the threshold, previously prepares a plurality of area-division pattern types, obtains affine transformations of each type starting from the type having the least number of divisions, calculates a prediction-error estimation value of each type, determines the suitable type which prediction-error estimation value is smaller than the threshold, applies a predictive frame by affine-transformation, encodes the prediction-error information and outputs the predictive frame together with side-information such as motion-vectors and area-dividing information.
|
