基于稀疏表示特征差异性的屏幕图像质量评估  

Quality Assessment for Screen Content Images Based on Sparse Representation Features Difference

作　　者：赵红梦 陈程立诏 ZHAO Hong-meng;CHEN Cheng-li-zhao(School of Computer Science and Technology,Qingdao University,Qingdao 266000,China)

机构地区：[1]青岛大学计算机科学技术学院

出　　处：《青岛大学学报（自然科学版）》2019年第4期66-71,78,共7页Journal of Qingdao University(Natural Science Edition)

基　　金：国家自然科学基金青年基金(批准号：61802215)资助

摘　　要：目前的屏幕图像质量评估方法大都是以单一级别的方式来评估图像的质量,这并不符合人类视觉系统的多层次特性。为此,提出了一种基于稀疏表示下梯度图像差异性的屏幕图像质量评估方法。将RGB图转为灰度图从而计算出梯度图后,学习参考图像字典,提取参考图像和失真图像的字典使用比重差异、长短差异、重叠率等特征,最后通过权重融合得到最终的屏幕图像质量评估结果。屏幕图像数据库(SIQAD)实验结果表明,该方法与人类主观分数有较高的一致性,在各类型失真图像的质量评估上也有优秀的表现。The current quality assessment methods for screen content images were mostly to evaluate the quality in a single level,which did not correspond to the multi-level characteristics of human visual system.To solve the problem,a method of screen content image quality assessment based on sparse representation features differences was proposed.The RGB image is transformed into gray-scale image and its gradient map was calculated.Then the reference image’s dictionary was learned and thus the dictionary usage atoms proportion difference,length difference and overlapping rate between reference and distorted version are extracted.The final evaluation result was obtained by weighted fusion of the above features.Experiments on screen content image dataset(SIQAD)showed that the proposed method was high consistent with human subjective scores and performs well in quality evaluation for various types of distorted images.

关 键 词：屏幕图像 图像质量评估 稀疏表示 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]