一种基于视觉感知的图像质量评价方法

A Picture Quality Evaluation Method Based on Human Perception

图像质量的好坏是评价图像压缩算法优劣的重要依据．传统的图像质量客观评价方法不能反映出图像在视觉感知上的失真，而主观评价方法如平均主观分数（ＭＯＳ）带有很强的不确定性．建立在人眼视觉模型（ＨＶＳ）基础上的感知均方误差（ＰＭＳＥ）做为主客观联系的桥梁，能有效地反映出人对图像失真在视觉上的感知．类似于均方误差（ＭＳＥ）的定义，我们采用二范数测度的失真测量来定义ＰＭＳＥ，并用ＷｅｂｅｒＦｅｃｈｎｅｒ衰减、调制传递函数（ＭＴＦ）过程和Ｂ样条小波进行模拟．我们对感知误差信号进行误差分割，对不同类型的误差进行不同的处理．最后去相关，加权和，得到一客观值，用于图像质量的客观评价．实验表明ＰＭＳＥ和ＭＯＳ具有很强的相关性．

Picture quality evaluation methods could be utilized to assess image algorithms′ performance.The traditional objective methods cannot necessarily represent the perception of error,and subjective methods such as Mean Opinion Score(MOS),extremely depends on observer′s experiences and motivations.Perceptual Mean Square Error(PMSE) is defined on the basis of Human Visual System(HVS) and then a model is developed to explain this definition.Three components such as Amplitude Nonlinearity,Modulation Transfer Function(MTF),Orientation and Scale Selectivity contribute to the output perceptual error.In them,B Spline shiftable and Multiscale Wavelet Transform is used to match eye′s Orientation and Scalability.An error segmentation process decomposes perceptual error to four sorts of errors which are then calculated by Mean Square Measurement.PMSE is thus the weighted sum of several principal components after the Principal Component Analysis.The model is tested on images finally,and the results show that PMSE correlates well with MOS.