基于视觉感知与学习的图像质量评价

Image quality assessment based on visual perception and learning

针对几种经典评价方法的缺点,通过引入人眼的主观特性,提出一种基于视觉感知与学习(visual perception and learning,VPL)的方法,以解决人类视觉系统(human vision system,HVS)多通道评价融合的不稳定性。借助反向传播(back propagation,BP)神经网络构建了评价融合模型,分别对几种视觉感知算法的多通道评价进行融合,并基于回归函数对视觉感知算法的结果进行二次互补融合。结果表明,相对于现有主流方法,本文所提方法的各项评价指标均具有较大的优势。

Aiming at the shortcoming of several classical assessment methods,a new method of image quality assessment based on visual perception and learning(VPL),was proposed by introducing subjective characteristics of human eyes with a view to solving instability of multi-channel assessment pooling of the human vision system(HVS).Then,an assessment pooling model was constructed by virtue of the back propagation(BP)neural network,pooling multi-channel assessment of several visual perception algorithms.Finally,second complementary pooling was conducted for results of every visual perception algorithm based on the regression function.The experiment results show that every assessment indicator of the proposed method has greater advantages compared to the existing prevailing methods.