基于视觉掩盖效应和奇异值分解的图像质量评测方法

图像质量的客观评测一直是视频和图像工程中富有挑战性的重要课题。传统的以PSNR为代表的检测方法对图像间的差异过度敏感，因此其结果常偏离主观感受；而以人眼视觉系统仿生理论为指导的评价方法则由于算法模型过于复杂而无法得到广泛应用。当前，基于图像特征参数检测的理论方法成为研究的新热点，设计能代表图像特征的参数是其关键。奇异值体现的是图像的内蕴信息，其对一定程度噪声扰动、比例伸缩、旋转、平移等具有较好的稳定性，因此非常适合作为图像的特征参数。本文通过将人眼视觉特性和图像本征特性相结合，设计一套基于视觉掩盖效应和奇异值分解的图像质量客观评测方法。该方法能在充分利用奇异值优势的前提下，根据视觉掩盖效应提高其对典型劣化形态的检测灵敏度，从而使评测结果与主观质量感受具有更好的相似性。

Image Quality Measure Based on Visual Masking Effect and Singular Value Decomposition

Image quality measure is one of the important parts of the video and image engineering. The objective quality assessment methods had experienced three phases. The first phase was characterized as MSE (mean square error) and PSNR (peak signal noise ratio) and so on. Their main theory was based on the by-pixel difference accounting, whose result could be sensible to the difference between two images. But sometime it was too sensible to have well correlation with the subjective one. The second phase was characterized as the PQS1] and JDM5] and so on. Their main theory was based on the simulation of the human visual system. But its model always too complicated to realize. Now it was more popular to measure the image quality by detecting the characteristic feature of an image. The singular value could always be the most important information of an image. It had well stability to some extent quality distortion such as noise, scale variety, rotation, clipping and so on. So it was suitable for singular value of an image to be used as the feature parameter. In this paper we combined the visual masking effect and singular value decomposition together and designed a model for objective quality measure of an image. The original and the distorted image were first pre-processed according to visual masking effect and then their image matrixes were transformed into vectors by singular value decomposition. By comparing the angle between singular vectors of them, the quality of an image can be measured. It could improve the sensibility of the traditional SVD method by taking the human visual system properties into account and its result had closer correlation with the subjective one.