基于决策优化和聚焦评价的多聚焦图像融合算法OA

To address the problems of blurred fusion edges and loss of details in the current multifocus image fusion,a multi-focus image fusion algorithm based on decision optimization and focusing evaluation is proposed.Firstly,the Sum of Modified Laplacian(SML)evaluation operator is chosen to generate the focusing evaluation map by focused measurement on the source image.Simultaneously,Convolutional Neural Network(CNN)is employed to perform feature learning on the source images,yielding a convolutional decision map.The decision map and focusing evaluation map are then used as labels to input into a random walk model,establishing the optimal transmission relationship between pixels to achieve boundary optimization of the fusion decision map.Finally,guided filter is used to further enhance edge consistency,and a weighted reconstruction is performed to generate the fused image.Experimental results on the Lytro dataset and captured images show that,compared with five advanced multi-focus fusion methods,the proposed algorithm achieves a maximum improvement of 0.022 in edge information retention Qabf and a maximum improvement of 0.282 dB in peak signal-to-noise ratio PSNR.The fused images exhibit superior performance in terms of edge details,contrast,and visual clarity,with a more natural fusion effect.