Samsung S5PC100 User Manual page 1577

S5PC100 USER'S MANUAL (REV1.0)
3.2.2 MOTION ESTIMATION
The Motion Estimation Block uses MRMCS (Multi-resolution search using multiple candidates and spatial
correlation of motion field) algorithm, and the search range is [+/-64, +/-32] pixel.
Motion estimation supports:
UMV (Unrestricted Motion Vector) mode
Up to quarter-pel search for H.264
Up to half-pel search for MPEG4
Supports 16x16/8x8 block for MPEG4
Supports 16x16/16x8/8x16/8x8 block for H.264
MRMCS algorithm
The Motion Estimation block has input as current pixels and pixels of search area, and calculates the optimal MV
(Motion vector) that has the smallest cost function between original MB pixels and reference MB pixels. To
calculate MVP (Motion vector prediction), ME block reads the upper MV from Line buffer memory and the left MV
from previous MB. After Motion Estimation, block type information, MVD (motion vector difference), predicted
pixels are transferred to the prediction block.
3.2.3 TRANSFORM & QUANTIZATION
Each residual macroblock is transformed, quantized and coded to remove the spatial redundancy. MPEG4 uses
8x8 DCT as the basic transform. H.264 encoder uses three transforms depending on the type of residual data that
is to be coded - transform for 4x4 array of luma DC coefficients in intra 16x16 macroblock, transform for 2x2 array
of chroma DC coefficients in any macroblock and transform for all other 4x4 blocks in the residual data. H.264
decoder supports 8x8 DCT and weighted quantization for high profile.
MPEG4 transform operates on 8x8 blocks of original data in intra macroblock or residual data after motion-
compensated prediction. MPEG4 decoder support weighted quantization for advanced simple profile.
H.264 transform operates on 4x4 blocks of residual data after motion-compensated prediction or Intra prediction.
The transform is based on the DCT but with some fundamental differences. Firstly, H.264 transform is integer
transform. Secondly, the core part of the transform is multiply-free, it only requires additions and shifts. Thirdly, a
scaling multiplication is integrated into the quantizer. The entire process of transform and quantization can be
carried out using 16-bits integer arithmetic and only a single multiply per coefficient without any loss of accuracy.
Transform and quantization IP does not use DMA. This IP operating frequency is under the 133MHz.
3.2.4 PREDICTION
Natural video image has both temporal and spatial redundancy. Therefore, through the prediction, coding
efficiency will be increased by transferring just residual for motion compression. For removing spatial redundancy,
H.264 standard additionally includes intra-prediction compared with previous standards. VME_PRED can operate
in Encoder or Decoder mode and each is composed of two parts. One is the INTRA PREDICTION part which
extracts predictors from neighbor pixels in same picture. The other is the INTER PREDICTION part which extracts
predictor that is interpolated by MV precision after being read from reference frames. When VME_PRED is
operated in decoder, decoder stores reconstructed pixels, which are derived from adding residual which is
MFC (MULTI FORMAT CODEC)
9.11-9