| 摘要 |
While accuracy of motion compensated prediction is kept high, reduction in processing amount is made possible. When motion compensated prediction is performed on inputted image data by a motion vector detection section, first, reference is made to past three reference frames stored in a reference image storage section, integer search is performed on the image data for each macroblock, and a motion prediction vector is detected by this. Based on the detection result of the motion prediction vector, a reference frame most appropriate for execution of motion compensated prediction is determined for each blocks. Next, based on the determination result, it is judged whether or not reference frames regarded as most appropriate for all blocks (nine blocks) are coincident to each other. As a result of the judgment, when the same reference frame is selected for all candidate blocks, only the reference frame regarded as most appropriate is made the object of sub-pel search, and the motion prediction vector is detected with sub-pel pixel accuracy.
|
