Research On Multi-focus Image Fusion Using Edge-preserving Filtering And Deep Learning

In recent years,with the large number of imaging devices used in the production and life of humans,different types of image data have become a commonly used information carrier in social life.However,depending on the imaging conditions of the imaging device,the obtained image data often cannot meet the actual application requirements.When the optical sensor captures images of the target scene,it is limited to the depth of field of the optical lens and cannot obtain clear images of the entire scene at one time.Therefore,the lens is usually focused for multiple acquisitions to obtain multiple source images focused on different regions.Integrating the clear information of these source images to produce a full-clear image is a problem to be solved in multi-focus image fusion.Multi-focus image fusion technique has high applicable value in various fields such as computer vision,medical imaging,satellite remote sensing,digital photography and many other applications.Under the above circumstance,this paper concentrates on the study of multi-focus image fusion algorithm based on edge-preserving filtering and deep learning.The main contents of this paper are listed as follows:1.Aiming at the problem that the sum of modified Laplacian for focus detection is insensitive to the homogeneous region,a new focus-measure called the sum of the weighted modified Laplacian for focus detection is proposed.Based on the modified Laplacian operator,the method sums the weighted values in the local window to increase the gradient contribution of the surrounding acute gradient to the region.This method takes the advantages of the sum of modified Laplacian sensitivity to gradients,and increases the focus detection performance on homogeneous region of the image.The experimental results show that the focus detection result of the method is more consistent with the source image,and the anti-noise ability is stronger,which is conducive to the further fusion rule design.2.By taking full advantage of the guided filter and using it in consistency verification,the fusion rule makes the algorithm achieve a more natural fusion effect in the boundary region.Experiments show that the proposed multi-focus image fusion method using SMML and guided filter is an effective image fusion algorithm and achieves a good fusion effect.3.By converting the focus detection into the binary classification problem of image blocks,the convolutional neural network has been applied to the focus detection method successfully.The method overcomes the problem that traditional single-image feature focus detection methods cannot be applied to complex and varied image scenes.Combining the fusion rules designed in this paper,the multi-focus image fusion algorithm based on deep learning achieves better performance in multiple scenes.