Classified directional prediction for visually lossless compression in display devices

This paper proposes a new visually lossless image compression technique for display devices. The algorithm explores the differences in error perceptibility of human visual system (HVS) for different image areas. Coding units are classified into smooth, detail and sharp areas based on local complexity and prediction errors. In order to achieve the objectives of imperceptible quality loss and high compression ratio, the classified blocks are quantized by different levels. The proposed architecture is designed to use only the memory of two line buffer, so it is hardware-friendly and could be easily implemented in many display devices. Experimental results show that the proposed method is visually lossless in all categories of tested images with the average compression ratio 6.138.