摘要
量化噪声是JPEG图像压缩过程中产生的主要失真,在图像质量评估时需要对其大小进行估计。针对峰值信噪比(PSNR)估计方法在实际应用时终端用户无法获取原始图像的限制,基于图像空域相关性,引入JPEG压缩图像的斜率均方差和最大非零离散余弦变换系数的位置作为可测特征量,利用训练图像进行机器学习得到训练模型,将测试图像进行解析得到图像PSNR盲估计结果。实验结果表明,该PSNR估计结果优于目前流行的峰值信噪比盲估计方法。
Quantization noise is one of thedominant distortions of JPEG image compression,so its amplitude usually needs to be estimated for image quality assessment. Meanwhile, Peak Signal to Noise Ratio (PSNR) estimation method cannot get the original image in the practical applications, so Mean Squared Difference Slope(MSDS) and the position of maximum nonzero Discrete Cosine Transform (DCT) coefficient are introduced as the measurable characteristics based on image spatial correlation. Some images as well as their PSNR are trained to produce the PSNR estimation model, and the test image is fed into the model to estimate its PSNR. Experimental results show that the proposed algorithm is superior to the existing blind PSNR estimation algorithms.
出处
《计算机工程》
CAS
CSCD
北大核心
2017年第8期253-257,共5页
Computer Engineering
基金
国家自然科学基金面上项目"基于内容的视频事件检测与描述研究"(61375016)
关键词
量化噪声
JPEG图像
峰值信噪比
图像相关性
斜率均方差
机器学习
quantization noise
JPEG image
Peak Signal to Noise Ratio (PSNR)
image correlation
Mean SquaredDifference Slope( MSDS )
machine learning