Research And Improvement On The Foreground Extraction Algorithms

Image matting technique, the process of extracting the foreground object from an image based on limited user input,is an essential problem for computer vision and has gained much attention in the field. In recent years, many approaches have been developed ranging from segmentation-based foreground extraction to image matting algorithms. This task of image matting is extremely challenging because it is ill-posed—at pixel level, many factors need to be estimated, including the foreground and the background colors, as well as the foreground opacity (alpha). In this paper, improvement work has been done for the closed-form solution to natural image matting after detailed research on the state-of-art techniques in relevant field. And the contribution of my work is mainly on the automatic realization for the image matting process based on the careful computation of the alpha value for edge details of the closed-form solution to natural image matting.The main work of this paper includes the analysis and survey about the traditional image extracting foreground technique: on the area of image segmentation, the edge detection, region segmentation and cluster segmentation are summarized, and experimental analysis is done of the Graph cuts algorithm based on the Graph theory; on the other side, several typical matting algorithms are discussed in detail together with their merits and demerits, and illustrative analysis is conducted on the fine Poisson matting algorithms. After that, the theoretical analysis and experimental verification of the closed-form solution to natural image matting algorithms are given for comparison with the Poisson matting method. Essentially the Color lines model of the corresponding histogram of the RGB space is built up for integration with the closed-form solution to natural image matting algorithm to yield improvement. Finally, the method's analysis is made and it proves to be an improved version through experiment.