Digital image matting is a classical topic in computer vision society and has a wide application in film production, graphic design and image editing societies. Thus has aroused more and more attention recently. It basically refers to the problem of extracting the region of interest such as a foreground object from an image based on user inputs like scribbles or trimap. More specifically, we need to accurately estimating the foreground color, background color, and an opacity (or transparency) value of each pixel for an input image. Matting is an ill-posed problem inherently since we need to output three images (foreground image, background image and alpha matte) out of only one input image. Therefore, in order to solve matting problem, various assumptions have been made to help...[ Read more ]

Digital image matting is a classical topic in computer vision society and has a wide application in film production, graphic design and image editing societies. Thus has aroused more and more attention recently. It basically refers to the problem of extracting the region of interest such as a foreground object from an image based on user inputs like scribbles or trimap. More specifically, we need to accurately estimating the foreground color, background color, and an opacity (or transparency) value of each pixel for an input image. Matting is an ill-posed problem inherently since we need to output three images (foreground image, background image and alpha matte) out of only one input image. Therefore, in order to solve matting problem, various assumptions have been made to help constrain it, and based on different assumptions, lots of matting algorithms were proposed. Nevertheless, natural image matting is still not an easy task for ideal alpha matte generating.

There are three main methodological parts in the thesis. Firstly, in order to gain more insights of matting problem, we start with a comprehensive survey and analysis of the existing matting literature. We observe that there are three key components in better estimating the alpha values, that is, the design of matting laplacian matrix, the definition of neighborhood and the choices of feature space. Based on this observation, we introduce a unified framework for digital image matting, which provides the possibility of obtaining a better understanding and direction of further improvement for image matting problem. The experimental results tested on different matting algorithms further prove the feasibility of our proposed framework.

Then, in the second part, inspired by closed-form matting and color clustering matting, we firstly develop an adaptive sample clustering criterion to automatically assign either local or nonlocal neighborhood to each pixel. After that, in order to enhance matting accuracy, we improve the nonlocal clustering performance by introducing a new feature selection parameter to choose preferred feature space for different images in a fully automatic way. And finally we solve the problem using a closed form solution. Experimental results show that our algorithm achieves equal or even better performance among many state-of-the-art matting techniques.

Apart from the hybrid method we adopted in the work proposed in the second part of our methodology, we find that there are still possibility to achieve better performance via a totally novel idea, that is, in all the conventional matting literature, people use compositing equation to describe the relation between original image and corresponding alpha matte. In contrast, in this part, we introduce a novel feature based compositing equation which encodes not only color information as in previous work but also coordinate information. Then, for the purpose of better preserving the intrinsic nonlocal structure of natural images, we propose an alpha-feature model based on the new compositing equation. Finally, we derive our matting method with closed-form solution achieved by using feature clustering ball model. Experimental results show that the proposed method outperforms most state-of-the-art matting methods.