Adaptive Block Compression Sensing Image Fusion

As we know,Image fusion has been widely used since it was proposed,especially in the field of national defense construction and economic development.With the increasing amount of acquired image data,the traditional pixel level image fusion method a huge pressure on data storage and transmission is unable to meet the requirements of big data anlysis.The theory of compressed sensing alleviates the problem of image processing with large amount of data,which could collect and compress the signal at the same time and reduce the amount of sampling data needed.However,the theory of block compressed sensing has obvious advantages in dealing with large-scale signal image.It has introduced the idea of block to sample and reconstruct the small image instead of the whole image.It can also solve the problem of large amount of sampling and reconstruction computation in compressed sensing.whereas,the block compression perception takes the same sampling rate for each image block and neglects the feature information contained in the image block,which may result in uneven distribution of sampling rate to get better reconstruction.So this paper would mainly focus on the application of the theory of block compressed sensing in image fusion,as a result an adaptive block compressed sensing image fusion algorithm has been proposed.And the main research work and achievements are as following:(1)Aiming at the problem of uneven distribution of sampling rates in traditional block compressed sensing,an adaptive sampling rate algorithm based on total variation is proposed in this paper.The total variation of image block is taken as the index to measure the best sampling rate of the block.When the overall sampling rate is fixed,the sampling rate is allocated based on the weight of the total variation of image block.As a result,the best sampling rate and reconstruction for each image block could be achieved by this algorithm under the same conditions.(2)As for image sparsity,the joint dictionary is used as the sparse base in this paper.The redundant dictionary obtained by this method has the structural characteristics of the original image,which can improve the sparse representation ability of the image.In addition,BCS?SPL(Block Compressed Sensing with Smoothed Projected Landweber)algorithm is applied in the reconstruction algorithm,which is able to eliminate the block effect caused by blocking.Meanwhile,Compared with other reconstruction algorithms,the advantages of this algorithm are also verified.(3)For different image blocks have different textures,this paper introduces the particle swarm optimization algorithm and takes the fusion coefficient of each image block as the particle.What's more the information entropy and average gradient of the measured value are taken as the optimization objective function.Eventually the optimal fusion coefficient of each block and optimal fusion of different texture blocks could be achieved by this algorithm.In addiction,the adaptive adjustment of fusion weights of different images and different texture features to some extent has been complished.(4)This paper has tested fusion algorithm with three types of traditional pixel level image fusion(IHS fusion algorithm,PCA fusion algorithm,Brovey fusion algorithm)and two kinds of non-improved fusion algorithm based on compression perception(CS?IHS fusion algorithm and BCS?IHS fusion algorithm).After the comparation and analysis of subjective visual effect and objective index evaluation,it's found that our algorithm does have great advantages in spectral fidelity,spatial resolution and detail texture.