Moulti-focus Image Fusion Algorithm In NSST Domain Based On Artificial Neural Network

Image fusion technology is the process of fusing two or more images into one image that contains more clear and effective information.The images of the same scene collected by different imaging devices are very different.At the same time,due to the limitation of the focal length of the same imaging device,a completely clear panoramic image cannot be presented.Therefore,the multi-focus image is used to process multiple multi-focus images with different sharpness in the same scene,and is used to obtain a full-focus image containing completely clear information.The processed multi-focus image is more consistent with the human-computer visual characteristics.So it is widely used in medical,military and other fields.There are three main strategies for multi-focus image fusion: pixel-level fusion,feature-level fusion,and decision-level fusion.The pixel-level image fusion method is a widely used type of fusion method,which is mainly divided into two types: spatial domain and transform domain.The image fusion method based on the transform domain performs multi-scale decomposition on the source images through multi-scale transformation.Then the image coefficients on different scales are fused using different fusion rules.However,such methods often cause distortion of the image.Multi-focus image fusion algorithms based on spatial domain are relatively simple and easy to operate,but such methods may produce block effects.With the rapid development of artificial neural networks,image fusion algorithms based on artificial neural networks have gradually gained widespread recognition from scholars in various countries.Aiming at the shortcomings of the existing algorithms,which is conducive to multi-scale analysis and the advantages of artificial neural networks,this paper proposes two methods based on spiking cortical model(SCM)and residual network(ResNet)combined with non-subsampled shearlet transform(NSST)multi-focus image fusion algorithm.Achieved a good image fusion effect.The main research of this article is as follows:(1)Multi-focus image fusion based on adaptive dual-channel SCM in NSSTIn order to alleviate the shortcomings of information loss in multi-focus image fusion images based on transform domain,a new adaptive dual-channel spiking cortical model(DualSCM)multi-focus image fusion algorithm based on NSST domain is proposed in thispaper.First,the basic fusion image is constructed in the NSST domain by registering the source images and DualSCM.After that,a focus area of the original input images based on the difference images between the basic fusion image and the original images are detected.Finally,the resulting fusion image is achieved by combining the focused areas.Due to the global coupling of DualSCM,this algorithm can well retain the information of the original images and obtain a fusion image that is more in line with human vision.(2)Multi-focus image fusion based on ResNet in NSST domainIn order to obtain a clearer panoramic image with more levels and texture information characteristics,a multi-focus image fusion algorithm based on ResNet in NSST domain is proposed.First,the source images are subjected to NSST decomposition to obtain high-frequency coefficients and low-frequency coefficients.Then,the generated low-frequency coefficients are fused through ResNet,and the generated high-frequency coefficients are fused using an improved gradient sum of Laplace energy(IGSML).Finally,the inverse NSST transform is performed on the fused high and low frequency coefficients to obtain the final fused image.When NSST processing is performed on the images,the low-frequency global features and high-frequency details of the images are fully considered.At the same time,ResNet has a deeper network structure,which can well obtain the spatial information characteristics of low-frequency coefficient images,which is conducive to the fusion of low-frequency coefficients.IGSML can use different directional gradients to process the information features of different levels and directions of high-frequency images,which is more conducive to the fusion of high-frequency coefficients.The experimental results show that the proposed method has improved the structural features,edge information and texture features of the fused image.