融合区域对比与稀疏低秩的显著性检测

Salient Detection Based on Region Contrast and Sparse Low-Rank

针对传统显著性检测方法在目标高亮和背景抑制等方面存在的问题,文中提出一种融合颜色区域对比与稀疏低秩的方法.采用超像素分割方法对原始图像进行分割,形成若干超像素块;结合图像颜色特征、纹理特征以及空间位置的独特性分别计算区域间的对比度,并对其进行融合得到初始显著图;利用稀疏低秩分解对初始显著图进行背景非显著信息的抑制,得到最终显著图.实验结果表明,文中方法能够均匀的突出显著目标,并有效抑制背景信息;在MSRA数据集上与其他6种方法进行对比实验,具有更好的准确率和召回率.

Existing saliency detection methods have the problems of target highlighting and background suppression.To solve these problems,the paper presents a saliency detection method based on region contrast and sparse low rank.The original image is segmented into superpixels.On the combination of the image color feature,texture feature and uniqueness of space location,the contrast between regions is computed,and the initial saliency map is obtained by fusion.The final saliency map is obtained by using sparse low-rank decomposition to suppress the background non salient information of initial saliency maps.The experimental results show that the proposed method can uniformly highlight the significant targets and effectively restrain the background information.Compared with the other 6 methods in the MSRA dataset,the proposed method is of higher accuracy and recall rate.