RGBD Saliency Object Detection via Regional Feature Clustering
Xiaojun Xia, Shuai Wang, Xiaodong Zhang, Lulu Cui, Zhiyang Zhao · 2019
According to the problem in the RGBD saliency detection, this paper proposes a RGBD saliency object detection algorithm based on regional feature clustering. First, the image segmentation is performed using a superpixel algorithm that uses depth information. Then the feature vector of each region is extracted. The region is clustered using ten different bandwidth MeanShift algorithms, and then obtain ten clustered maps. Next, the ten clustered maps are calculated to generate ten saliency maps. Ten saliency maps are merged into one saliency map with a XGBoost model. Then the saliency map is added as a new feature to the feature vector mentioned above. Continue to calculate the saliency map until the loop is up to ten times to output the final saliency map. Experimental results on three public RGBD datasets show that the algorithm has better performance.