H.264编码下基于DCT系数量化值的PSNR估计方法  

H.264 coding PSNR estimation method based on quantized DCT coefficients

作　　者：沙亮[1] 王贵锦[1] 赵安邦 林行刚[1] 唐昆[1] 

机构地区：[1]清华大学电子工程系,北京100084

出　　处：《清华大学学报（自然科学版）》2008年第4期510-513,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60472028);国家教育部博士点基金项目(20040003015)

摘　　要：针对广泛用于测量视频信号质量的指标峰值信噪比(PSNR)因需源信号而受限的问题,提出一种无需参考源的PSNR估计方法。该方法从视频码流中提取离散余弦变换(DCT)系数的量化值进行统计,得到其概率密度函数,并以此依据Laplace模型估计PSNR。该方法针对最新的视频压缩标准H.264的帧结构、变换和量化结合方式做出了优化,使得不同的编码速率和帧类型的码流均可适应。使用不同码率和不同运动方式的H.264码流进行测试,比较使用新方法得到的PSNR与实际PSNR间的差异。结果显示CIF格式的中低码率情况下二者的均值在1 dB以内。A peak signal-to-noise ratio （PSNR） estimation algorithm that does not use a reference source was developed to address restrictions caused by source signals in calculating the PSNR, which is a widely used index for video quality. The algorithm enables PSNR estimates based on the Laplacian probability density function of the quantized discrete cosine transform （DCT） coefficients extracted from a video bit stream. The algorithm is optimized for the video coding standard H. 264 based on its frame structure and its combination of transform and quantization, which accommodates different bit rates and frame types. Tests on various rates and motion videos give average difference of 1dB between the estimated PSNR and the real PSNR for moderate bit rates.

关 键 词：离散余弦变换(DCT)系数 LAPLACE分布 H.264 峰值信噪比(PSNR) 量化 

分 类 号：TN919.81[电子电信—通信与信息系统]