A Comprehensive Study on Image Matting Techniques for Alpha Estimation

Natural Image Matting (NIM) possesses toughest challenge in the sector of computer graphics and computer vision. The problem is poorly presented when only input image is given, with no further information. Image matting is the technique of softly fetching a foreground object from the background in an image according to the definition provided by the matting equation. Based on their techniques of estimating alpha-matte, image matting methods can be broadly classified into Sampling based, Propagation based and Learning based matting methods. In this paper we have done a comprehensive analysis on various image matting methods along with their advantages, disadvantages and challenges. Their mean square error (MSE) and sum of absolute differences (SAD) parameters are also analyzed. In this analysis we can conclude that Learning based methods outperforms other two methods. Among Learning based methods Cascade Picture Matting Network with Deformable Graph Refinement (Cas-DGR) and Tripartite Information Mining and Integration Network (TIMI-Net) approaches give better performance than other methods based on their MSE and SAD analysis.