图像依赖的显著图融合方法

Image-dependent fusion method for saliency maps

提出基于脊回归的显著图融合方法以获得更好的检测效果。在训练集中寻找待检测图像的近邻图像集,对近邻图像集采用脊回归方法对多种显著性检测方法的融合系数进行估计,进而对不同检测方法的显著图进行融合。该方法充分考虑了检测方法的差异性,很好的解决检测图像在没有基准二值标注下显著图的融合问题。试验采用流行的显著性数据集和显著性检测方法,本研究方法在ECSSD数据集上的AUC为0.911,在HKU-IS数据集上的AUC为0.987,在DUT-OMRON数据集上的AUC为0.953,结果验证了融合方法的有效性。

A saliency fusion method based on ridge regression was proposed to obtain better detection performance. The nearest neighbor set of the image to be detected was searched in the training set. The ridge regression method was used to estimate the fusion coefficients of different saliency maps. The saliency maps of different detection methods were fused. This method fully considered the differences of detection methods, and solved the problem of saliency map fusion in the absence of benchmark binary annotations. The AUC value of the proposed method was 0.911 on ECSSD dataset. The AUC value of the proposed method was 0.987 on HKU-IS dataset. The AUC value of the proposed method was 0.953 on DUT-OMRON dataset. The efficiency of the proposed method was verified by experimental results.