摘要： 针对采用区域间对比度检测图像显著性带来的涂抹效应的不足,提出一种超像素显著性检测算法。采用简单的线性迭代聚类超像素分割方法将图像分割成大小不同的区域。根据区域间对比度检测图像显著性区域,运用自适应中心环绕对不同分割尺度的显著图进行处理,消除显著图的涂抹效应,通过多尺度融合得到最终显著图。在数据集MSRA-1000上对该算法进行验证,结果表明,相比SEG,MSS等算法,该算法具有更高的查准率与召回率。

关键词: 显著性检测, 自适应中心环绕, 多尺度融合, 超像素分割, 块效应

Abstract: Aiming at the deficiency of blockiness produced by using the contrast between regions for saliency detection,a multi-scale superpixel saliency detection algorithm is proposed.The algorithm uses a Simple Linear Iterative Clustering(SLIC) superpixel segmentation method to divide an image into regions of different sizes,and detects image saliency map according to the contrast between regions.An adaptive center surranding method is proposed to eliminate the blockiness of the salienly map.The final saliency map is obtained by multi-scale fusion.Experimental results on data sets MSRA-1000 show that,compared with classic algorithms like SEG,MSS and so on,the algorithm has higher precision and better recall rate.

Key words: saliency detection, adaptive center-surround, multi-scale fusion, superpixels segmentation, blocking effect

中图分类号: 

• TP391