基于背景信息优化的显著性目标检测

Salient Object Detection Based on Background Information Optimization

摘要: 自然图像中的背景包含了丰富的信息，在进行显著性目标检测时，如果处理不当，会影响检测结果的准确性。为了减少背景对检测的影响，本文结合背景信息特征提出了一种相邻像素优化的图像显著性目标检测方法。首先提取背景的统计信息和结构信息构建初始的显著性图；然后，利用保边平滑滤波器对初始显著图进行优化，加强目标区域的细节信息，获得轮廓较为清楚的显著图；最后，根据相邻像素之间的影响，建立相邻像素之间的关系对显著图做进一步优化，得到最终显著图。在两种公开的数据集上测试，并与四种经典的显著性检测算法对比，采用精确率–召回率曲线和F-measure图对算法进行评估，结果显示，本文提出的算法生成的显著图效果更好，检测的准确性更高。

Abstract: The rich background information of natural images will affect the accuracy of the results in the pro-cess of salient object detection. In order to reduce the influence of background, we propose a salient object detection method based on adjacent pixel optimization, which exploits the feature of background information. Firstly, an initial saliency map is generated by exploiting statistical and struc-tural information of the background. Secondly, in order to enhance the detailed information of the object region and obtain a saliency map with sharp edge, the initial saliency map is optimized by using an edge-preserving filter. Finally, the relationship between adjacent pixels is established to further optimize the saliency map, according to the influence of adjacent pixels. Experiments on two public datasets and compared with four mainstream detection methods, we use the precision-recall rate curve and F-measure graph to evaluate the algorithm. The results show that the saliency map generated by the proposed method in this paper is better than other methods and more effective in highlighting the salient object uniformly.