Image Matting Algorithm Research Based On Optimized Global Sampling

With the rapid development of computer software and hardware level and information technology in the last few years, digital image matting technology is widely applied in image processing, video editing and special effects making of film, which is an important research in the computer vision field.Matting is a technology that accurately extracting the foreground objects from an input image according to different applications. Because the matting problem is highly ill-posed, some user interaction is necessary to add constrains and make the problem solvable. Trimap and strokes are common interaction modes for matting. The current matting technology can be mainly categorized into blue matting and natural image matting, in which natural image matting can be further divided into four categories: sampling-based approaches, propagation-based approaches, energy function-based approaches and some extension ones.In this dissertation, we present an optimized global sampling algorithm after deeply analyzing and comparing the existing algorithms. Our method treats all foreground and background pixels on the boundary as candidate samples. For each unknown pixel, the algorithm can select (approximately) global optimal sample pairs from candidate sample set by a valid sample evaluation criteria and sampling mechanism and estimate its value of foreground opacity. The advantage of our method is that can deal well with the relation between nearby samples and distant samples and avoid leaving good samples out. In the phase of optimizing result, we introduce Global Optimization Method and Guided Filter Method and compare the two methods. Besides, we accomplish a method of generating trimap by use of image segmentation and morphological image processing, which facilitate the user interaction. Finally, all of the related algorithms are accomplished in the experiment platform and the images in the testing set are processed. The experimental results show that our method can produce high-quality mattes.