A Deep Interactive Segmentation Method with User Interaction-based Attention Module and Polar Transformation

摘要

Interactive segmentation that extracts a specific foreground selected by the user input is widely employed in manyuser-interactive applications such as image editing and ground-truth labeling. In general, most interactive segmentationmethods iteratively refine the previously obtained result using additional user interactions because they often produceunsatisfactory results with a single user input. A recently developed convolutional neural network (CNN)-basedinteractive segmentation method called deep interactive object selection has achieved high segmentation accuracy withfewer user interactions than earlier non-CNN-based approaches. However, the computational efficiency of deepinteractive object selection deteriorates due to the repetitive feature extraction stage for each user interaction.Furthermore, the deep interactive object selection requires graph cut as a post-processing step to refine the boundarysegments. To solve this problem, this paper presents a deep CNN-based interactive segmentation method employing aneffective and simple user interaction-based attention module that does not require the repetitive feature extraction. Inaddition, we adopt Cartesian to polar coordinate transformation to further improve the segmentation performance.Experimental results demonstrate that the proposed interactive segmentation method is superior to the conventional onesin terms of segmentation accuracy and computational efficiency.