Research On Multi-focus Image Fusion Algorithm Based On Deep Learning

Image fusion technology integrates two or more images that have different features in the same scene into one picture,so that it contains more valuable information in each source image.Among them,multi-focus image fusion is an important branch: many photos at different focus points in the same location have different clear parts,and the clear parts of multiple photos are merged into one picture,which is conducive to further research on image processing.Activity level measurement and fusion rules are two key factors in image fusion.For most of the existing fusion methods,both in the spatial domain or in the transform domain such as wavelet,the activity level measurement is basically achieved by designing a local filter to extract high-frequency details.Fusion rules usually apply different rules to different frequency band coefficients.However,in order to achieve satisfactory fusion performance,these two tasks are usually difficult to complete at the same time.In the study of multi-focus image fusion,this paper solves this problem with the deep learning method,which simultaneously completes the feature extraction and fusion criteria of the multi-focus image and generates a fusion image that is more in line with the human visual system.The effective training of deep neural networks requires a large amount of labeled training data.However,there is no available publicly labeled data set for academic research at the present stage in the field of multi-focus image fusion.Besides that,manual data calibration takes a lot of time and efforts and the quality is difficult to be uniform and stable.Based on the above considerations,this paper studies the following main contents:First,a data set of different objects are used to identify real scenes in VOC 2012,and a network training set is obtained by setting a mask method.Secondly,using the generalized LFCNN of the convolutional neural network on the image intensity channel Y to perform the fusion calculation,which overcome the difficulties in the separate design of feature extraction and fusion criteria in the traditional fusion algorithm.Based on the above considerations,this paper proposes a new multi-focus image fusion method based on deep learning.Experimental results show that the proposed method can overcome the disadvantages of traditional algorithm effectively,and our method has greatly improved the visual quality and objective evaluation compared with the traditional algorithms.