Filter for eliminating block

Abstract

The de-blocking filter is base on interlaced filter sequence and uses two 1D filter units combined with macro block data memory unit, upper part data memory unit, left data memory unit and intermediate data memory unit so as to greatly increase the filter speed. By selecting proper filter sequence, the invention utilizes the data-to-data dependency; the data of each block required by the filter can be read once from outside so as to reduce the access times to the external SRAM / SDRAM and to reduce the complexity of filter operation; meanwhile, a parallel storage mechanism is used in the memorysubsystem and data path in order to simultaneously support the parallel access in both vertical and horizontal directions. By introducing the memory structure provide by the invention, the traditional transposition circuit is not needed any more.

Description

【Technical field】 [0001] The invention relates to a deblocking filter for video encoding and decoding, in particular to a deblocking filter based on the H.264 / MPEG-4 AVC video standard. 【Background technique】 [0002] H.264 / MPEG-4 AVC is the latest video standard formulated by Joint Video Team (JVT). This new standard has made significant improvements to eliminate the virtual boundaries between blocks, and greatly improved the image quality. However, in this standard, the image will appear blocky after the codec inversely transforms and quantizes the image. There are two reasons, the most important reason is the discrete cosine transform (DCT) based on the intra-frame and inter-frame prediction residual of the block, because the quantization process of the transform coefficient is relatively rough, so the transform coefficient restored by the inverse quantization process With errors, it will cause visual discontinuities on image block boundaries. The second reason comes fr...

AI Technical Summary

Problems solved by technology

There are two reasons, the most important reason is the discrete cosine transform (DCT) based on the intra-frame and inter-frame prediction residual of the block, because the quantization process of the transform coefficient is relatively rough, so the transform coefficient restored by the inverse quantization process With error, it will cause visual discontinuity on the image block boundary

The second reason is from motion compensated prediction, where motion compensated blocks may be copied from interpolated sample data at different positions in different frames

However, the deblocking filter is in the loop, which requires a large amount of calculation and data access. It easily occupies more than one-third of the computational complexity in the entire decoder. The filtering speed of the filter affects the speed of the entire codec. Therefore it is necessary to use a fast filter

[0004] The current deblocking filter structure mostly uses a single boundary filter unit to support both horizontal filtering and vertical filtering operations, which greatly limits the speed of deblocking filtering and affects the speed of the entire encoding and decoding process.

In addition, the traditional deblocking filter uses a transpose circuit to deal with the access to column data in the process of horizontal boundary filtering, so that the rate of vertical filtering and horizontal filtering is almost doubled, which is not conducive to parallel access to two directions at the same time, and Also increases the circuit area

Images