Pixel-Copy Prediction Based Lossless Reference Frame Compression.

With the increasing demands of high definition and high resolution for video applications, the bandwidth and power consumption of accessing external memory storing reference frames during motion estimation bring serious pressure on practical video coding systems. Lossless reference frame compression is a proper method to decrease memory size and access bandwidth without any quality loss. This paper proposed a pixel-copy prediction based lossless reference frame compression. The method predicts current pixel by copying adjacent reconstructed pixels adaptively based on the estimations of the differences between the original pixel and the left and upper reconstructed samples. Then residuals are encoded by Huffman encoding to generate bit stream. Realized with HEVC reference software HM-16.5, experimental results show that our method achieves averagely 67.45% data reduction ratio (DRR) for luminance component that outperforms pervious works on computation complexity and compression efficiency.