An Efficient Lossless Image Compression Algorithm for External Memory Bandwidth Saving

Summary form only given. Huge SDRAM memory access bandwidth is the performance bottleneck for high definition video coding. Lossless image compression is efficient method to solve this problem. Image pixels are compressed before writing into SDRAM and decompressed after reading out from SDRAM. This work proposes a hardware-oriented lossless image compression algorithm, supporting block and line random access patterns flexibly. First, block or pixel level adaptive prediction is employed to utilize the image spatial correlation. Rate optimized mode decision is used to select the block or pixel prediction mode, and to determine the horizontal or vertical prediction direction, to achieves good tradeoff between prediction accuracy and control overhead cost. Second, multiple range semi-fixed variable length coding is employed for residue coding, and Huffman based VLC tables are designed to code the prediction mode syntax elements, as shown in Fig.1. The proposed algorithm and its competitors are tested with 35 standard test images with size of 960x576. The average bits of the proposed algorithm, algorithm [1], algorithm [2] and CALIC algorithm are respectively 3.79, 3.95, 4.19 and 3.92 bits per pixel. The results justify the effectiveness and its advantage of the proposed algorithm.