SReN swMATH ID: 28348 Software Authors: Z. He, Y. Zhou, Y. Wang, Z. Tang Description: SReN: Shape Regression Network for Comic Storyboard Extraction. The goal of storyboard extraction is to decompose the comicimage into storyboards, which is the fundamental step ofcomic image understanding and producing digital comic doc-uments suitable for mobile reading. Most of existing ap-proaches are based on hand crafted low-level visual patterslike edge segments and line segments, which do not capturehigh-level vision information. To overcome this drawbackof the existing approaches, we propose a novel architecturebased on deep convolutional neural network, named ShapeRegression Network (SReN), to detect storyboards withincomic images. Firstly, we use Fast R-CNN to generate rect-angle bounding boxes as storyboard proposals. Then we traina deep neural network to predict quadrangles for these pro-posals. Unlike existing object detection methods which onlyoutput rectangle bounding boxes, SReN can produce moreprecise quadrangle bounding boxes. Experimental results on7382 comic pages, demonstrate that SReN outperforms thestate-of-the-art methods by more than10 Homepage: http://philokey.github.io/sren.html Related Software: FASText; COCO-Text; R2CNN; TextProposals; TextBoxes; PVANet; EAST; Characterness; TextBoxes++; SSD; Adam Cited in: 1 Publication Cited by 3 Authors 1 Bai, Xiang 1 Liao, Minghui 1 Shi, Baoguang Cited in 1 Serial 1 IEEE Transactions on Image Processing Cited in 1 Field 1 Information and communication theory, circuits (94-XX) Citations by Year