一种自适应空间邻域的显著图获取方法研究

摘要

In order to improve the situation of low accuracy of image retrieval in complex backgrounds,this paper proposed a retrieval scheme based on adaptive circular spatial surround.Considering the concentric structure of the human eye reception field,it adopted circular spatial surround.Then it calculated the salient values of pixels in that surround through weights generated by a two-dimensional normal distribution.Finally,it employed a simple threshold segmentation process on the saliency to generate a binary image.Experiments were carried out on 2 natural image sets.The results show that,compared with three state-of-the-art salient region detection methods,the proposed method gives improved accuracy in acquiring the saliency.%针对复杂背景下显著图提取精准度不高的问题,提出了一套基于自适应空间邻域的获取方案.该方案考虑人眼神经元感受野的同心圆结构,计算自适应圆形空间邻域；然后结合二维正态分布的显著权值计算空间邻域内每个像素点的显著值,获取图像的显著图,再利用简单的阈值分割算法提取二值图像；最后通过在两个自然图像集进行实验,并与三种经典算法进行比较.实验结果表明,该方法可以在复杂背景下有效地获取精确的显著图.