侯春萍,李浩,岳广辉.局部和全局特征融合的色调映射图像质量评价[J].湖南大学学报:自然科学版,2019,(8):132~140
局部和全局特征融合的色调映射图像质量评价
Quality Assessment of Tone-mapped Images Using Local and Global Features
  
DOI:
中文关键词:  图像质量评价  人类视觉系统  色调映射  无参考
英文关键词:Image Quality Assessment(IQA)  Human Visual System(HVS)  tone mapping  No-reference(NR)
基金项目:
作者单位
侯春萍,李浩,岳广辉 (天津大学 电气自动化与信息工程学院天津 300072) 
摘要点击次数: 25
全文下载次数: 26
中文摘要:
      人类视觉系统首先粗略地感知全局区域,然后精细地感知局部区域的图像质量.针对色调映射图像的质量评价问题,考虑人眼视觉机制的特性,提出一种融合局部和全局特征的无参考图像质量评价算法.首先从全局特征出发,考虑了颜色矩、全局熵和欠曝光/过曝光条件下的明暗分布特性,得到相应的全局特征;然后结合局部对比度、局部熵和分块小波能量,得到相应的局部特征;最后,融合全局特征和局部特征,使用支持向量回归进行特征训练,建立图像特征空间与感观质量分数的关系,得到图像质量评价模型.在公开的ESPL-LIVE HDR数据库上验证,实验结果表明,提出的方法与主观评分有较高的一致性,并且性能优于目前较优秀的无参考图像质量评价算法.
英文摘要:
      Human Visual System(HVS) first roughly perceives global areas,then centers on the detailed local areas for the perception of image quality. In this paper,a novel blind Image Quality Assessment(IQA) algorithm was proposed for tone-mapped images by combining local and global features. First,the global features were extracted based on color moments,global entropy and bright/dark pixels' distribution under overexposure/underexposure conditions. Then,local contrast,local entropy and wavelet energy based on blocks were utilized to extract local features. Finally,global features were combined with local features to constitute a final feature vector. And all these feature vectors mentioned above were trained using Support Vector Regression(SVR) to generate a model,which bridges the feature space with quality space. Extensive experiments on a public ESPL-LIVE HDR database have demonstrated that the proposed method has a high consistency with subjective evaluation and outperforms state-of-the-art no-reference IQA metrics.
查看全文  查看/发表评论  下载PDF阅读器
关闭