Quick automatic head image matting method based on segmentation and propagation

Image matting is a process of extracting objects from background in an image, which is an important work in digital image processing and video editing. Previous methods have poor performance and most methods need trimap or scribble to compute to get accurate image matting results. In this paper a quick automatic head image matting method for certificate photo production is put forward, which could rapidly extract satisfied head photo from images that are shoot by portable camera. First a new training data set about hair matting is created, and the image is segmented into different regions according to different gray level; then we detect and locate the face and eyes to adjust the correct head position; Finally, the accurate hair pixels are extracted from the edge region around head by multimodal Gaussian process regression. It's the advantage that the regions with clear foreground and background could be quickly extracted by segmented method, and the regions with similar foreground and background colors or complicated textures were labeled (the edge regions with hair around head), thus the hair could be extracted in smaller area. In our method the quick and accurate extracting of head photo needn't the trimap or scribbling or lots of training. Experimental results clearly demonstrate the superiority of our algorithm over previous methods. (C) 2019 Elsevier B.V. All rights reserved.