Method of scalable fractional motion estimation for multimedia coding system
A multimedia coding system has different picture qualities with different coding characteristics. It is done through a scaleable fractional motion estimation (FME) in H.264 and a block size trend prediction FME. Three quality levels are thus obtained. The first level processes a complete calculation and obtains a best picture; the second level, fewer calculation with a low power consumption and an acceptable picture quality; and the third level, fewest calculation with a fastest speed. And the present invention is suitable for hardware design to obtain a high efficiency, a low cost and a high performance.
1. A method of a scalable fractional motion estimation (FME) for a multimedia coding system, comprising the steps of:
(a) processing an H.264 coding to obtain a quality scalability of three quality levels and processing a scalable FME to said three quality levels; and
(b) processing a block size trend prediction FME (BSTP-FME) to estimate required area and parallelism through one or more search ranges and obtaining a best selection according to signal-to-noise ratio (SNR) drop wherein said BSTP-FME comprises:
controlling mode flow by receiving integer motion vectors (IMVs) to decide a flow for said IMVs;
calculating motion vectors (MVs) by receiving said IMVs and processing motion estimations to corresponding fractional pixels to obtain MV costs;
dual-directionally processing rotation interpolations to a plurality of the IMVs to obtain calculation data;
calculating a sum-of-absolute-transformed-differences (SATD) by receiving said calculation data and obtaining SATDs of said corresponding fractional pixels; and
selecting cost by receiving said MV costs and said SATDs to be accumulated and compared to obtain a best parallelism of hardware processing.
2. The method according to claim 1 ,
wherein said quality levels are obtained by dividing results of FME into three quality levels according to characteristics required by respective usage levels, comprising:
(i) a single mode, said single mode processing an integer motion vector (IMV);
(ii) a concise mode, said concise mode processing a plurality of IMVs to obtain a better picture quality than that obtained through said single mode; and
(iii) a complete mode, said complete mode processing IMVs of all block modes to obtain a best picture quality.
3. The method according to claim 1 , wherein said one or more search ranges are included in a range of ±2 fractional pixels with a full search to process motion estimations to 25 fractional pixels.
4. The method according to claim 1 , wherein said calculating MVs comprises referring to a reference table.
5. The method according to claim 1 , wherein said calculating the SATD comprises processing 20 processing elements (PE) based on 4×4 macroblocks.