基于图像能量相关性的H.264/AVC 2阶预测算

Second-order Prediction Algorithm Based on Image Energy Correlation for H.264/AVC

针对现有2阶预测算法对残差像素间的相关性考察不够,使得2阶预测不够准确,导致编码效率降低等缺点,提出一种基于图像能量相关性的H.264/AVC 2阶预测算法。该算法基于对帧内、帧间残差像素相关性的详细统计分析,定义了像素能量相关性,通过统计出能量相关性和1阶帧间残差的关系曲线,动态地、较为准确地筛选出2阶帧内预测参考像素,既提高了2阶帧内预测的准确性,在码率降低的同时,提高了编码图像质量。实验结果表明,本文算法在图像质量和码率两方面均优于现有2阶预测算法。尤其对于高分辨率且包含复杂运动的视频,图像质量和码率提升较大,与其它2阶预测算法相比,率失真优化效果明显。

Based on image energy correlation, a novel second-order prediction algorithm was proposed to solve the low coding efficiency caused by inaccurate prediction and no detailed analysis of the correlation between the pixels. By analyzing the statistical properties be- tween intra residual and inter residual, the energy correlation between pixels was defined. Then, the statistics of the energy correlation and first-order inter residual curve were applied to filter out second intra prediction reference pixels dynamically and the second-order prediction became more accurate which makes a bitrate reduction and the coded image quality improvement. Experimental results showed that the bitrate and the quality of image are significantly improved. Especially for high-resolution video with complex movement, this algorithm achieves significant PSNR increment and bit rate reduce on average. The rate-distortion optimization has a significantly improvement.