Multi-focus image fusion using learning based matting with sum of the Gaussian-based modified Laplacian

Abstract

In most traditional multi-focus image fusion algorithms, the focus measures used to detect the focused areas of multiple images are unstable and sensitive to noise. Although many modified methods implement more sophisticated strategies to cope with this problem, the complexity of these methods turns out to be a problem for mobile devices. In order to cope with these problems, a novel multi-focus image fusion method based on image matting technique is put forward in this paper. This method is simple yet effective. Firstly, a new focus measure named the sum of the Gaussian-based modified Laplacian (SGML) is proposed to estimate the focus map of source images. Then, the initial segmentation map can be obtained via a novel sliding window strategy. Further, with the rough segmentation map as an input, learning based image matting technique is performed to extract the exact boundaries between focused and defocused regions. Finally, by combining the information of the focus areas, an all-in-focus image can be obtained. The numerous experiments have revealed that the proposed approach yielded better results in comparison with some competing techniques both in subjective and objective evaluation.