Enhancement of segmentation and saliency image by using Multiple instance learning
Paper ID : 1096-MVIP2020
omid mohammadi *1, Mohammad Javad Fadaei Eslam2
1Semnan University, Faculty of Electrical and Computer Engineering/Master student
2Assistant Professor of Electrical and Computer Engineering Department/ Semnan University, Iran
While both co-segmentation and co-saliency methods can achieve common objects and salient areas separately, they also have the potential to exchange information as a stand-alone method and complement each other. In this paper, we try to improve the accuracy of results compared to previous work in this field using multiple learning methods. The challenges to be considered are first the complexities of the background, the similarity of the background, and the common object, which is resolved by the optimization method in the multiple learning, and then the output different changes in size, brightness, rotation, etc. We use insurance to extract deep features. To extract the feature we use pre-trained networks and apply high-level features from the bottom layers of the network. Finally, we apply our approach to the problem of minimizing the energy function of a segmentation graph. The results of the proposed approach on the benchmark dataset show the superiority of this approach over previous approaches.
co-Segmentation, co-saliency, multiple learning, deep features, energy function
Status : Paper Accepted (Poster Presentation)