多特性融合的多尺度显著性检测

Multi-scale Saliency Detection Via Multi-feature Fusion

图像显著性检测已经发展多年,被广泛应用于目标检测与识别、图像分割等领域中.本文针对图像显著性目标和背景相似时,检测结果较差的问题,提出一种多特性融合的多尺度检测方法,有效改善了这个情况.方法主要有三个阶段:首先构建图像层次的多尺度图像;其次在单尺度图像上,融合基于对象性、背景性和外观性的检测结果,获得单尺度显著性图;最后融合多个单尺度显著性图获得最终的显著性图.在两个公开数据集ASD和ECSSD上,将本文方法同其它12种流行的显著性检测方法进行比较,结果表明本文方法优于其它方法,能够更加准确地确定显著性目标区域,尤其在图像显著性目标和背景相似的情况下,也能保证较高的准确率.

Saliency detection for images has been studied for many years. Salient object detection is widely used in object detection and recognition,image segmentation,etc. But it does not work well when the image traget is similar to the background. In this paper,a multi-scale and multi-feature fusion detection method is proposed to solve the problem. The proposed method consists of three basic steps. Firstly,image-level multi-scale images are constructed;Secondly,on the basis of the single-scale image,the detection results obtained using objectness,background characteristic and appearance characteristic are fused to produce the single-scale saliency map. Finally,all the single-scale saliency maps are merged to obtain the final saliency map. We compare the proposed method with other popular saliency detection methods on two public datasets,including ASD and ECSSD. The experimental results show that the proposed method is superior to other methods and can more accurately determine the salient object region,especially in the case where the image target is similar to the background,which can also achieve higher accuracy.