Bust Portraits Matting Based on Improved U-Net

Extracting complete portrait foregrounds from natural images is widely used in image editing and high-definition map generation. When making maps, it often necessary to matte passers-by guarantee their privacy. Current matting methods that do not require additional trimap inputs suffer inaccurate global predictions or blurred local details. Portrait matting, as a soft segmentation method, allows the creation of excess areas during segmentation, which inevitably leads noise resulting alpha well foreground information, so keep all areas. To overcome above problems, this paper designed contour sharpness refining network (CSRN) modifies weight values uncertain regions prediction map. It combined with an end-to-end for bust based on U-Net target detection containing Residual U-blocks. An designed. The can effectively reduce without affecting information obtained by deeper network, thus obtaining more detailed fine edge structure has been tested PPM-100, RealWorldPortrait-636, self-built dataset, showing excellent performance both refinement half-figure portraits.