Abstract:
The distribution of coefficients of image in the DCT domain can be modeled by a Laplace probability density function with parameter λ. Quantization noise of a compressed image can be estimated from its quantization parameter and λ. This paper proposes a method to estimate the distribution parameter λ and Peak Signal to Noise Ratio(PSNR) according to trained image blocks in spatial domain. As no original image is required, this method is actually no-reference PSNR estimation.
Key words:
Laplace distribution,
compressed image,
quantization noise,
Peak Signal to Noise Ratio(PSNR)
摘要: 经过离散余弦变换的图像在DCT域系数的分布近似符合一个用参数λ描述的拉普拉斯分布。利用该参数以及图像在JPEG压缩中使用的DCT域量化系数,可以实现对图像量化噪声的估计。提出一种基于图像先验知识的分布参数估计方法,可以在没有未压缩的原始图像作为参考时实现对λ值的估计,进而计算压缩图像的峰值信噪比。
关键词:
拉普拉斯分布,
压缩图像,
量化噪声,
峰值信噪比
CLC Number:
DONG Hao-Yuan, FANG Xiang-Zhong, TUN Zhi-Kai. Blind Estimation Algorithm for Quantization Noise Based on Image Priori Knowledge[J]. Computer Engineering, 2010, 36(11): 195-197.
董皓远, 方向忠, 吴智恺. 基于图像先验知识的量化噪声盲估计算法[J]. 计算机工程, 2010, 36(11): 195-197.