Object Extraction Algorithm Based on Saliency Prior Information

Abstract

Object extraction provides the basis for subsequent tasks such as object recognition, and its importance is self-evident. For this purpose, this paper proposes an object extraction algorithm based on saliency prior information. Firstly, the SIFT operator and the oriented edge forest method are used to extract the saliency points and the saliency edges respectively. Then construct a simple small convolutional neural network for the saliency fusion task, and fuse the saliency point and the saliency edge to obtain the saliency fusion map. Then the fusion map is added to the object extraction network structure as a priori information. In this paper, the residual network is used for the object extraction task, and finally the high-quality object extraction work is realized. The algorithm uses IOU parameters and Precision as the quantitative evaluation index, and uses qualitative analysis to comprehensively analyse the experimental results. Experiments show that the proposed algorithm has good results in both image quality and visual, and has good robustness.