Research On The J-Linkage Clustering With Applications To Image Matting

Image matting is a hot topic in the research of computer vision. To meet the increasing requirements from applications, the foreground extraction technique has developed from image segmentation to high level alpha value computation. So far, image matting has already witnessed applications in realms of segmentation, compression and color editing and saturation correction. Specifically, image matting is an ill-posed problem, and for each image pixel we have to estimate the alpha values of the foreground area, the background area and the transitional area, making the given task challenging.We propose an approach for image matting using J-Linkage clustering, and simplify the user supervision process into merely selecting pixels that represent the main colors of the foreground. In this approach, the J-Linkage clustering algorithm is used to model the color lines with geometrical parameters of spatial lines, and these lines are grouped by the seed information of pixels selected by users, as foreground lines and background lines respectively. During this process, these pixels produce seed information to yield a tri-map and we use the closed-from solution to finally acquire a matte.The state-of-the-art matting algorithms require user supervision as constraints to yield ideal matting results, by adding commonly a tri-map or several strokes. The merits of our algorithm can be summarized as building color lines model efficiently using J-Linkage and reducing user intervention tremendously to just several clicks on the image, making the matting procecss more automatic. Essentially, our approach avoids the potential failures caused by coarse tri-map classification and improper stroke scribbling. Experimental results demonstrate the robustness of our approach.