A Multi-Focus Image Method Based on Gradientlet Filter and Convolutional Sparse Representation

In order to address the issue with low contrast in multi-focus image fusion, this paper exhibits a multi-focus image fusion technique based on convolutional sparse representation (CSR) and gradient filtering. Firstly, the preprocessed multi-focus source image is decomposed by gradient filter, which is divided into approximation layer and residual layer. Secondly, for the approximation layer, CSR fusion rules are used for fusion. Then, for the residual layer, the improved parameter adaptive PCNN fusion rule is used for fusion. Finally, the final fusion result map is obtained by gradient filter reconstruction. Based on the experimental results, it can be concluded that this method of fusion produces better results than the other five methods in terms of avoiding ghosting at the edge of the target object and preserving the brightness, edge, detail, and texture information of the source image. The five objective evaluation indexes of standard deviation (STD), spatial frequency (SF), average gradient (AG), directional gradient energy (sobel8_grad), and visual fidelity (VIFF) reflect this improvement.