A Image Fusion Method Based On Shift-Invariant Shearlet Transfom And Deep Convolutional Neural Network

Image fusion is to synthesize the image or image sequence information about a specific scene acquired by two or more sensors at the same time or at different times to generate a new information processing process about the interpretation of this scene,which is not available from the information acquired by a single sensor.In recent years,researchers at home and abroad have made fruitful achievements in the field of image fusion,This paper will focus on the image fusion method under translation invariant shear wave transform.Compared with other traditional multi-scale transformation methods,SIST is the most advanced multi-scale analysis method at present,which can realize the "sparse" representation of images and efficiently extract useful information of images.In addition,unlike the traditional shear wave transform,SIST does not downsample the source image in multi-scale decomposition,so it has the translation invariant property,can retain the details well and overcome the pseudo-Gibbs effect.How to fuse the sub-bands obtained by SIST decomposition of the source image has become the focus of the research,In this paper,deep convolution neural network and adaptive pulse coupled neural network are used to improve the effect of sub-band fusion.The specific research contents of this paper are as follows:1)This paper summarizes the research status of image fusion at home and abroad,and elaborates the evaluation criteria of image fusion performance.Then,in each chapter,the concepts and related theories of shear wave,guide filter,depth convolution neural network,scale invariant feature transformation and pulse coupled neural network are introduced.2)In order to improve the quality of image fusion,an image fusion method based on translation invariant shear wave transform and adaptive impulse coupled neural network fusion rules is proposed.The SIST is used to perform shear wave decomposition on the source image,and the low-frequency components obtained by the decomposition are subjected to low-frequency fusion using the fusion rules based on the image-guided filter,and the improved spatial frequency is used as the PCNN input for the high-frequency components,and the improved Laplacian is used.The energy of the sigma and the link strength as PCNN,and finally the fused image is obtained by SIST inverse transform.3)In order to overcome the problem of traditional image fusion methods,an image fusion method based on the shift-invariant shearlet transfom and deep convolutional neural network is proposed.Firstly,the high frequency components obtained by SIST decomposition are fused by CNN fusion rules;Next,scale invariant feature transformation is used to find feature descriptors for low-frequency components,and the location of each feature descriptor inlow-frequency subbands is recorded,Then,the fusion strategy based on matching degree is used for low-frequency fusion;Finally,the fused image is obtained by inverse SIST transform of the high-frequency and low-frequency components.The experimental results show that compared with the classical traditional fusion method,the proposed method has a significant improvement in both subjective and objective evaluation indicators.4)Aiming at the image enhancement and registration before fusion,this paper designs and implements a computer-aided fusion platform.In each module,image color reversal,binarization,Laplacian sharpening,image smoothing and edge detection are realized respectively,and human-computer interaction is carried out in a simple and intuitive visual form.