Research And Realization For Image Compositing Methods Based On Image Matting And Adaptive Color Preservation

With the continuous development of the field of image and multimedia information, a variety of commercial image processing software are gradually increasing. Nowadays, the research on more efficient compositing and editing methods for various kinds of images has been a hot topic in the field of image analysis and processing.In this paper, the research contents mainly include the foreground extraction for complex images, the effective reduction of color distortion and less human interactions in the image compositing field. Firstly, for the compositing problem of boundary seam caused by the inaccurate foreground extracting masks for complex images, the research of matting for the foreground extraction has been highlighted in this paper. Because most of the current studies of matting focus on the global accuracy for the alpha matte estimation, thus leads to the large local error. In this paper, upon the matte with global accuracy by using the comprehensive sampling matting method, local similarity for the matte estimation has also been taken into account, and the initial matte is further optimized locally and iteratively. In this way, a more accurate mask for the compositing both on the global and local can be achieved. Experimental results show that the accuracy of masks by the matting method in this paper is improved both for the benchmark datasets and the additional shadow image datasets.Then, taking into account of the limitations of the existing two categories of compositing methods(gradient-based and alpha blending), we propose a hybrid hierarchical weighted compositing method by combining this two kinds of methods. That is, divide the blending regions according to the alpha matte by matting first, and then conduct the hierarchical compositing task in these different regions. In order to improve the efficiency of our algorithm, the gradient-based method is implemented with MVC(mean-value coordinates). And in order to realize the seamless compositing as well as preserve the original color of the source image, an innovative idea of the adaptive color fidelity weight by intensity sampling for the balance adjustment is put forward. Experiments show that our proposed method is more effective both for normal images and shadow ones compared with the existing methods.Finally, we build an easy-operating image compositing system, which can provide multiple functions and modules, including user input, foreground extraction, image compositing, results demonstration and comparison, etc. With the system, the effective compositing between the source and target image can be achieved quickly. Study in this paper shows the exploration and attempt for the foreground extraction problem of complex images(such as shadow images), as well as the adaptive, fast and efficient image compositing methods. It also shows the potential application value for the improvement and enhancement of more superior image processing and editing software.