Abstract: Existing algorithms are less effective in detecting images with complex background. A novel salient detection algorithm is proposed by using region contrast and background prior. Firstly, the image is segmented into perceptually homogeneous patches. Then region contrast is employed to obtain global contrast and spatial distribution maps. And then, background prior is adopted to compute the similarity between the patches and background. Finally three feature maps are fused into a saliency map based on superpixels. To better exploit each pixel’s color and position information, the fused saliency map is refined. Experiments on two popular benchmark datasets demonstrate that the proposed approach can uniformly highlight the salient object and effectively suppress the background.

Key words: salient object detection, saliency, background prior, region contrast

摘要： 针对现有算法对复杂背景的图像检测效果较差的问题，提出融合区域对比度和背景先验的显著目标检测算法。首先利用超像素分割将图像分割成感知均匀的图像块，然后根据区域对比度计算全局对比度特征和空间聚集度特征，再根据背景先验得到背景集，计算图像块与背景集间的相似性特征，接着对三个特征显著图进行融合计算，最后根据每个像素与周围超像素的颜色和距离对比度得到每个像素的显著值。实验结果表明，所提算法能较均匀高亮整个目标且有效抑制无关背景信息。

关键词: 显著目标检测, 显著性, 背景先验, 区域对比度