摘要
图像质量评价(Image Quality Assessment,IQA)是计算机视觉领域研究的基本问题之一。目前,绝大多数图像质量模型都是基于灰度图像构建的,而彩色图像质量评价至今依然是IQA领域的开放性问题。彩色图像质量评价研究的关键在于建立与人类色彩认知能力相一致的色彩信息的量化描述。本文基于颜色名称(Colornames,CN)构建彩色图像质量评价模型,将图像的每个像素值映射为CN概率向量,利用Wasserstein距离计算两幅图像的感知色差,以亮度和梯度特征作为补充,在池化阶段采用显著性加权得到客观图像质量评分。在公开测试数据集上的实验结果表明,提出的模型在TID2008、TID2013和最新的KADID-10k数据集上表现最佳,其SROCC值分别为0.9009,0.8901,0.8637。总体评价效果与目前最好的传统方法(非深度学习方法)相当;而对于颜色失真,则具有明显的优势。
Image quality assessment(IQA)is one of the basic research issues in the field of computer vision.At present,most image quality models are constructed based on grayscale images,and color image quality assessment is still an open issue in the field of IQA.The key of color image quality assessment research is to construct a quantitative description of color information consistent with human color cognition.This paper constructs a color image quality assessment model based on colornames(CN).It maps each pixel value of the image to a CN probability vector,uses the Wasserstein distance to calculate the perceived color difference of two images,uses the lightness and gradient features,and uses the saliency weighting in the pooling stage to obtain the objective image quality scores.The experimental results on the public test databases show that the proposed model performs best on TID2008,TID2013 and the latest KADID-10k databases,with SROCC values of 0.9009,0.8901 and 0.8637,respectively.The overall assessment effect is comparable to the current best traditional method(Non-deep learning method).But for color distortion,it has obvious advantages.
作者
马畅
张选德
MA Chang;ZHANG Xuan-de(College of Electronic Information and Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, China)
出处
《液晶与显示》
CAS
CSCD
北大核心
2022年第1期56-65,共10页
Chinese Journal of Liquid Crystals and Displays
基金
国家自然科学基金(No.61871260)。
