基于稀疏性和自信息的显著性检测方法  

Fast saliency detection method based on sparsity and self-information

作　　者：刘尚旺[1,2] 赵胜杰[1,2] 李元隆[1,2] 

机构地区：[1]河南师范大学计算机与信息工程学院,河南新乡453007 [2]河南师范大学"智慧商务与物联网技术"河南省工程实验室,河南新乡453007

出　　处：《计算机工程与设计》2016年第8期2176-2180,2194,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(U1304607);河南省重点研究基金项目(15A520080);河南师范大学博士启动基金项目(qd12138)

摘　　要：为提高图像显著性检测的准确性与有效性,在CIE Lab颜色空间内,通过模拟生物视觉神经元的中央-周围运算,提出一种基于稀疏表示与自信息的快速显著性检测方法。对原始输入图像的特征图像进行稀疏量化,计算该稀疏量化图像各像素点的自信息值,根据各像素点的自信息值进行显著性检测。实验结果表明,与GBVS、AIM和ITTI模型相比,该方法 AUC值分别提高了15%、17%和20%,平均耗时则分别降低了93%、95%和92%,验证了该方法能够准确快速地检测图像显著性。To increase the accuracy and efficiency of image saliency detection,through mimicking the center-surround（C-S）operations of biological vision neuron,a fast saliency detection method based on sparsity and self-information was proposed in the CIE Lab color space.The image features of original input image were sparsely quantified,and the self-information of each pixel in the sparse matrix was calculated.Subsequently,saliency detection was implemented according to the self-information of each pixel.Experimental results show that,comparing with the GBVS,AIM and ITTI models,the areas under the receiver operating characteristic curve（AUC）of the proposed method increase 15%,17% and 20%,respectively.Correspondingly,the average time-consuming decreases 93%,95% and 92%,respectively.The proposed method can detect the saliency more accurately and efficiently.

关 键 词：稀疏表示 自信息 显著性检测 颜色空间 中央-周围运算 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]