边界先验和自适应区域合并的显著性检测  

Saliency detection based on boundary prior and adaptive region merging

作　　者：翟继友[1,2] 屠立忠[1] 庄严[1] 

机构地区：[1]南京工程学院计算机工程学院,南京211167 [2]河海大学计算机与信息学院,南京211100

出　　处：《计算机工程与应用》2018年第6期178-182,共5页Computer Engineering and Applications

基　　金：南京工程学院校级科研基金(No.CKJA201306)

摘　　要：为了对图像中的显著目标进行更精确的识别,提出了一种基于边界先验和自适应区域合并的显著性检测算法。采用超像素分割算法对图像进行过分割,把超像素看做图的一个顶点来进行构图;定位和消除错误边界,使背景基准集中存在很少的噪声,减小目标接触图像边界时造成的误检;采用单通道索引颜色直方图度量区域相似度并进行区域合并得到显著图。对比实验表明该算法相比其他算法取得了较高的查准率,说明了算法的有效性。In order to efficiently extract the salient region of images, a novel saliency detection algorithm based on boundary and adaptive region merging is proposed. Firstly, the input image is over-segmented by using a superpixel segmentation algorithm and a graph is constructed by taking a superpixel as a vertex. Then through locating and eliminating error bounds, there is little noise in the background, and errors are reduced when objects touch the boundary of images. Finally,single channel index color histogram is used to measure the regional similarity and saliency map is obtained by region merging. Experimental result shows that this algorithm has higher precision than other algorithms and shows the effectiveness.

关 键 词：边界 自适应 区域合并 显著性检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]