基于多尺度多特征的显著性区域检测

Salient region detection via multi-scale and multi-feature

针对目前大多数显著性检测方法显著性区域轮廓不明显的缺点,提出了一种基于多尺度多特征的显著性区域检测算法.该算法利用基于改进的八邻域算法和基于熵率的超像素分割方法,获取到不同尺度下图像的亮度特征、颜色特征和纹理特征,从而使显著性区域的轮廓更明显.该算法在MSRA-1000、ECSSD、THUR15K 3个公开数据集上进行实验,并与现有的8种算法(FT,AC,IT,GB,AIM,SEG,SIM,SUN)做了对比,实验结果表明,该方法能够有效地提高图像的检测效果.

In view of the disadvantages that the contour of the saliency area is not clear in many salient detection methods,we proposed a detection method based on multi-scale and multi-feature for saliency detection. We used the improved eight neighborhood method and the entropy-based super-pixel segmentation method to obtain the brightness characteristics, color features and texture features of the images at various scales so that the contours of the significant regions are more obvious. The algorithm was tested on three public datasets,MSRA-1000,ECSSD, and THUR15 K, compared with eight existing algorithms（FT, AC,IT,GB,AIM,SEG,SIM,SUN）. The experimental results show that the proposed method can improve the detection effect of the image effectively.