基于稀疏系数和FV向量的图像质量评价  

Image quality assessment based on sparse coefficient and FV vector

作　　者：李博文 范赐恩[1] 颜佳[1] 冯天鹏 

机构地区：[1]武汉大学电子信息学院,湖北武汉430072

出　　处：《计算机工程与设计》2018年第3期763-768,共6页Computer Engineering and Design

基　　金：国家自然科学基金青年科学基金项目(61501334)

摘　　要：为降低基于码本的图像质量评价算法的训练时间,有效保留系数矩阵各个维度的特征信息,提出基于稀疏系数和Fisher向量的无参考图像质量评价算法。采用显著性图像块建立稀疏编码字典,减小冗余图像块节省训练时间,通过Fisher向量对稀疏系数编码,进入支撑向量回归预测图像质量。实验结果表明,小尺寸稀疏编码字典可以替代大字典,提取的特征能在无参考的情况下更好地评价图像质量。To reduce the training time of codebook-based image quality assessment methods and effectively retain comprehensive feature information on all dimensions,a metric based on sparse coefficient and Fisher vector for no-reference image quality assessment named SCAFV was proposed.Salient image patches were used to build the dictionary of sparse coding and it was computationally appealed by reducing highly redundant learning patches.Fisher vector was used to encode sparse coefficients.Support vector regression was performed to predict the image quality scores.Experimental results show that small size dictionary of sparse coding can replace large size dictionary and the final features yield superior performance in the absence of reference.

关 键 词：无参考图像质量评价 稀疏编码 稀疏系数 Fisher向量 支撑向量回归 

分 类 号：TN911.73[电子电信—通信与信息系统]