Abstract:
In traditional multi-focus image fusion methods, decision maps produced by focus measurement are frequently sensitive to noise and misregistration, and are prone to identification errors such as burrs, small holes, and small isolated areas in the focus detection area. A multi-focus image fusion algorithm based on multi-scale morphological focus measurement and optimization of random walk is proposed for the above-mentioned problems. First, multi-focus image generates the initial decision map through multi-scale morphological focus measurement, multi-scale morphological focus measurement has high focus area detection accuracy and can well recognize the contour of the image. The focus area in the decision map is then reconstructed using morphological filter and small block filter, removing the burrs and small isolated areas in the initial decision map. Using the optimized random walk algorithm to model from the perspective of probability, by solving an alternative objective function to estimate the probability of each pixel in the decision map. Finally, the optimized random walk algorithm is used to model from the perspective of probability, by solving an alternative objective function to estimate the probability of each pixel in the decision map associated with the observed pixel, to generate the optimal decision map. The experimental results show that in the subjective evaluation, the proposed method has a clear and artifact-free local magnification image and is able to align the boundaries better. In the objective evaluation, the proposed method achieves significant advantages in all eight indicators.