Multi-focus Image Fusion Based On Super-resolution Reconstruction

Due to the depth of field of optical lens is limited,when there is an object in the shooting scene that is not in the range of depth of field,the object outside the range of depth of field will be blurred.As a result,the photos taken partial blurring and partial clarity,which cannot provide high-quality pictures for computer vision and other applications.Now in order to solve this problem,We use a zoom cameras to capture images which in the same scene with different focal lengths,and using multi focus image fusion technology to fuse these image gathered into one full-focus image,it makes the objects in the fusion image are clear,and to provide the high-quality images for the subsequent application.At present,multi-focus image fusion technology plays an increasingly important role in medical imaging,reconnaissance equipment,information security and other aspects.The multi-focus image fusion algorithm is mainly divided into the fusion method based on transform domain and the fusion method based on spatial domain.The methods based on transform domain include pyramid transform,wavelet transform and multi-scale geometric analysis.The transform based on spatial domain mainly includes pixel-based fusion method,block-based fusion method and region-based fusion method.In recent years,due to the rapid development of neural networks and the fact that the neural network-based image fusion methods are more robust than the traditional multi-focus image fusion methods,CNN,GAN and other methods have been widely used in the multi-focus image fusion technology.Most of these multi-focus fusion technologies based on neural networks rely on external data sets or trained models.In this paper,the multi-focus image fusion method is studied.In order to solve the problems existing in the multi-focus image fusion,the following researches are carried out:(1)A multi-focus image fusion algorithm based on super-resolution reconstruction is proposed.Firstly,the algorithm uses ZSSR to reconstruct two multi-focus images with super resolution respectively,then two reconstructed images are obtained.Secondly,the structural similarity of the multi-focus image and the reconstructed image is calculated respectively.Thirdly,the decision graph of fusion image is obtained by comparing two structural similarity matrices.Finally,small area coverage and morphological processing were used to optimize the decision graph,and two multi-focus images were fused by the decision graph to obtain a full-focus image.Through the contrast experiment of multi-focus image fusion,it is found that this method has better fusion effect and the fused image quality is higher.(2)A multi-focus image fusion algorithm based on image matting is proposed.In the algorithm,the weighted average fusion of two multi-focus images is carried out to obtain the fused image,and the super-resolution reconstruction(ZSSR)is carried out to obtain the reconstructed image.Secondly,it is assumed that the reconstructed image and two multi-focus images conform to the local linear function relationship,the coefficient of the linear function is obtained according to the local linear model.and the parameters ? of the matting model are calculated according to the coefficient of the obtained linear function.Then,use the 8-neighborhood method to find the connected region with small area,and using the small area coverage method to optimize the parameters,so that the parameters matrix can clearly distinguish the focus region and the defocus region in the image.Finally,after smoothing the edge of the parameters matrix ? with guided filtering,two multi-focus images are fused by the parameters matrix ? to get the fully focused image.In the experiment,Firstly,contrast experiment was carried out on the algorithm parameters,and select the optimal parameter value.Then the comparative experiment results of multi-focus image fusion show that this method can better distinguish the focal region from the defocused region,reduce the unsmoothness of the transition region,and improve the quality of the fusion image.The two multi-focus image fusion algorithms proposed in this paper are stable and reliable,and have good processing effects on noisy images.They can provide ideas for other multi-focus image fusion methods and promote the development of multi-focus image fusion technology.