Research on Interactive Image Segmentation based on Differential Click Network

• 1、School of Computer Science,Beijing University of Posts and Telecommunications, 100876

Abstract：Depth image segmentation technology plays an important role in many fields, such as image restoration, image understanding, content retrieval, video segmentation and so on, but there are some problems, such as obvious boundary segmentation, occlusion error segmentation and so on. Interactive image segmentation technology gives users a lot of operating space, imagination and freedom to make up for the shortcomings of automatic image segmentation technology. The interactive image segmentation based on click form takes the mainstream deep learning network as the main architecture, the input of the network is the fusion input of RGB image and interactive data, and the user\'s click is continuously input until satisfactory results are obtained. In previous studies, all clicks are treated equally, while perceptual consideration of different click interaction information plays different roles. In this paper, a DCN (Differentiated Click Network, differential click network model is proposed to separate and extract feature units from user interaction data to enhance the information weight of different regions of the image, so as to improve the segmentation accuracy of the model. The effectiveness of DCN model has been fully verified in ResNet34, ResNet50 and MobileNet network architecture, and good results have been achieved in GrabCut, SBD and Berkeley open datasets and wx, wxsod and aliwenyu datasets.

Keywords： Interactive image segmentation Differentiated Click Network Convolution neural network