Improved BING Method and Its Application in Object Detection

Abstract

Abstract-In this paper, we proposed an improved saliency computing method based on BING method. We observed that the undetected objects of BING method have something in common-that is, most of them are occluded or truncated. Therefore, we improved BING method by: Firstly, make a new training set with undetected objects (by BING method) and truncated ground truth of the original training set. Then, train an assistant filter on this new training set. The assistant filter supplements BING method by detecting objects which BING method misses successfully. The experimental results show that the detection rate with improved BING method is increased from 97.2% to 98.1% for 2000 proposals, and that our method, with training of an assistant filter, is better than original BING method at finding incomplete objects.