自适应阈值分割与局部背景线索结合的显著性检测

为了提高显著性算法对不同类图像的适用性以及结果的完整性，该文提出一种基于自适应阈值合并的分割过程与新的背景选择方法相结合的显著性检测算法。在分割过程中，生成相邻区块的RGB以及LAB共六通道融合的颜色差值序列，采用区块面积参数的反比例模型生成自适应阈值与颜色差值序列进行对比合并。在背景选择过程中，根据局部区域背景-主体-背景的相对位置关系线索，得到背景区域，再对结果进行边缘优化。该算法与其它算法相比得到的显著图不需要外接其他阈值算法即生成二值图，自适应阈值合并能排除复杂环境中的物体细节，专注于同等级大小物体的显著性对比。

Saliency Detection Based on Adaptive Threshold Segmentation and Local Background Clues

In order to improve the applicability for different types of image and integrity of the results, a saliency detection algorithm is proposed. It combines the adaptive threshold merging with a new background selection strategy. In the segmentation process, the color difference sequence is obtained by the selective fusion of RGB and LAB of adjacent blocks. Adaptive threshold is generated by inverse proportion model of block area parameter. Merging progress is done after the adaptive threshold comparison with the color difference sequence. In the background selection process, background regions are obtained by the local relative position of background-subject-background in the local area. The experimental results are optimized for edge. Compared with other algorithms, the saliency map of two values obtained does not need external threshold algorithm in this paper. Adaptive threshold merging can eliminate the details of objects in complex environments and can focus on the saliency comparison of the same level size objects.