Research On Image Fusion Based On Improved Sparse Representation And Neural Network

Image fusion technology is an important research in image processing.The purpose of image fusion is to fuse information from multiple sensors about multiple images acquired by the same object,to extract the most useful information from multiple images,and to fuse it into a new image.The fused image not only retains the important feature information of the source images,but also enhances the clarity of source images,improves the quality of the source image.In addition,it can be more suitable to human visual system and convenient for the computer to do the next processing.At present,the main research directions of image fusion technology include the medical image fusion,infrared and visible image fusion,medical image fusion and so on.It has been widely used in medical research,remote sensing processing,geological prospecting and military target recognition.Based on the image fusion technology and the characteristics of multi-scale analysis tools,this paper studies the source image fusion and proposes two image fusion algorithms as follows.Firstly,an image fusion algorithm based on improved sparse representation and PCNN(Pulse Coupled Neural Networks)is proposed.In the beginning,the proposed algorithm decompose source images with NSST(Non-Subsampled Shearlet Transform),and it attains a low-frequency sub-band and several high-frequency sub-bands with different scales and directions.For the low-frequency sub-band,the proposed algorithm uses the improved sparse representation as the fusion rule,the K-SVD algorithm is used to find the adaptive learning dictionary,and construct multiple sub-dictionaries into a joint dictionary.The proposed algorithm uses the orthogonal matching pursuit(Orthogonal Matching Pursuit,OMP)algorithm to get the sparse coefficients,and finally obtains the fusion coefficients of the low-frequency sub-band information of the source image.For the high-frequency sub-bands,the proposed algorithm uses the improved pulse coupled neural network as the fusion rule,the improved spatial frequency is used as the feedback input of PCNN neurons,and the high-frequency sub-band coefficients of source image are chosen according to the fusion rule of the maximum total amplitude of ignition output.Finally,the NSST transforms fused low-frequency sub-band coefficients and high-frequency sub-band coefficients to obtain the fused image inversely.The experimental result shows that the proposed algorithm can preserve the feature information of the source images,and the fusion image is clearer than the source images,moreover,the proposed algorithm also achieves good results in objective evaluation criteria.Secondly,an image fusion algorithm based on improved guided filtering and dual-channel PCNN is proposed.In the beginning,the algorithm decomposes the source image at multi-scale to obtain a low-frequency sub-band and multiple high-frequency sub-bands with NSST.For the low-frequency sub-band part of the image,the proposed algorithm uses the improved guided filtering algorithm as the fusion rule.In order to enhance the image edge information,we use the phase consistency as the weight factor of guiding filter.For the high-frequency sub-bands of the image,the proposed algorithm uses the improved dual-channel PCNN as the fusion rule,the proposed algorithm uses the improved sum of Laplacian energy as the excitation input of the dual-channel PCNN,and the improved spatial frequency is used as the link strength.Finally,the fusion image is obtained by inverse transform of NSST.The experimental results show that the algorithm has advantages in image fusion effect and fusion rate,and also improves the quality and accuracy of the fused image.