Multifocus Image Fusion Based On Non-subsampled Shearlet Transform

Image fusion is to extract and fuse the effective information from multiple image sensors to get a complete image with more information.Multi-focus image fusion is an important branch of image fusion.Due to the limited depth of field of optical sensor,it is difficult to obtain all the images focused at the same time.Multi focus image fusion can fuse multiple images and get all the images with clear scenes.In this paper,multi-focus image fusion algorithm is based on nonsubsampled shearlet transform(NSST),the main research work includes the following two parts:(1)In view of the problems of unclear focus,edge blur and ghosting in traditional multi focus image fusion algorithms,difficult parameter setting of traditional PCNN,and sparse representation can not fully represent the feature information of the source image.A multi-focus image fusion algorithm based on NSST domain parameter adaptive PCNN and convolutional sparse representation is proposed.NSST was used to decompose the source image at high and low frequencies,and the parameter adaptive PCNN model was introduced into the fusion of high frequency coefficients,which solved the difficulty of manually setting parameters.Convolutional sparse representation is applied to low-frequency subband fusion to solve the problem of sparse representation's insufficient ability to save edge details.The experimental results show that the algorithm proposed in this chapter has good performance in both subjective visual effect and objective evaluation.(2)In order to further improve the quality of multi-focus image fusion,the problem that the multi-focus image fusion algorithm based on convolutional neural network does not distinguish the high and low frequency of the image during image fusion is solved,and the fusion results do not conform to human visual perception.A multi-focus image fusion algorithm combining NSST and convolutional neural network is proposed.On the basis of obtaining the high and low frequency coefficients of the source image decomposed by NSST,the initial weight map of the source image was obtained by using the convolutional neural network,and the initial weight map was decomposed into the high and low frequency components.Different fusion rules were used to fuse the high and low frequency coefficients of the source image and the high and low frequency components of the initial weight image.Finally,the final fusion result was obtained by NSST reconstruction.The experimental results show that the proposed algorithm can effectively express the details of the image,make the details of the fused image clearer,and make the edges smoother.It has the ability to retain useful information of the source image and capture the deeper geometric structure of the source image.