基于视觉感知高度相关的图像质量评价

Image quality evaluation based on high correlation of visual perception

评估并监控图像质量是数字图像处理技术的基础工作。客观图像质量评价(IQA)旨在通过计算机开发与人眼视觉感知密切相关的算法。本文充分模拟人眼视觉系统(HVS)和大脑机制,提出了一种新的基于机器学习的全参考型图像质量评价模型CGDR。该模型融合了图像的色度特征、梯度特征、对比敏感度函数(CSF)特征以及Gauss差分(DOG)频带特征。其中,改进后的梯度算法不仅包含更丰富的相邻信息和多方向边缘信息,同时强调了参考图像和失真图像的边缘相关性。在三个基准图像数据库上的实验结果表明,CGDR的预测性能优于八种主流方法,跨数据库测试体现出其强大的鲁棒性,预测结果能够与人眼主观感知保持高度一致性。

Evaluating and monitoring image quality is the basic work of digital image processing technology.The objective image quality assessment(IQA)aims to develop algorithms closely related to human visual perception.This paper fully simulates the human visual system(HVS)and brain mechanism,with a new machine learning-based full reference image quality assessment(FR-IQA)model CGDR proposed.This model combines the chrominance features,gradient features,contrast sensitivity function(CSF)features and difference of Gaussian(DOG)band features of the image.Among them,the improved gradient algorithm not only contains richer adjacent information and multi-directional edge information,but also emphasizes the edge correlation between reference image and its distorted version.The experimental results from the three benchmark image databases show that the proposed method has a better prediction performance than the current eight mainstream approaches.The cross-database validation shows its strong robustness,which is highly consistent with the human subjective perception.