基于颜色和空间融合的显著图算法

A saliency map algorithm based on color contrast and region connectivity

针对计算机视觉模型中显著图求解的问题,本文提出了基于颜色和空间融合的显著图算法。首先对图像进行超像素分割,再根据图像颜色对比度及空间信息,计算颜色显著图和空间显著图,最后根据区域连通性和动态阈值自动去除非显著区域,完成显著性目标计算。将本文算法在MSRA10K_Imgs_GT数据集上得到的结果和几种主流算法进行比较,通过人眼观察以及准确率、召回率、F-measure值指标评判,结果表明本文算法的显著性目标计算更为准确。

This paper researched on the problem of saliency map in computer vision model, we proposed an automatic saliency region detection algorithm based on Color and Space Fusion. Firstly, the image is segmented, then according to the image color contrast and spatial information, the color saliency map and spatial saliency map are calculated. Finally, the non-salient regions are automatically removed based on regional connectivity and dynamic threshold. The results obtained from the MSRA10K_Imgs_GT data set and several main algorithms are compared. The results show that the proposed algorithm is more accurate and more accurate.