A hybrid defocused region segmentation approach using image matting

摘要

In this paper, a hybrid defocused region segmentation using image matting is proposed. The technique incorporates three sharpness metrics which are magnitude spectrum slope, local total variation and local binary patterns to identify the in-focus pixels in the image. Trimap is generated automatically using sharpness maps to obtain the prior information and matting Laplacian is applied to propagate the trimap to the entire image based on color similarities. Simulation results compared using visual and quantitative metrics show the strength of the proposed technique.