Research On Compressed Sensing Image Fusion Algorithm Based On Regional Characteristics

Image fusion is an important branch of information fusion,it is the multiple sensors sampling and collecting on the same scene or different scenes of the same target multiple source images for appropriate fusion processing,in order to obtain more clear,full,rich images of the same scene or object.It is widely used in remote sensing detection,traffic monitoring,medical image analysis,environmental protection and other important fields.With the extension of its application,the number of images in the process of image fusion is gradually increasing,and the enormous amount of computation gives the image fusion process a great difficulty.Compressed sensing theory provides a new and convenient way to solve this complex problem.When the signal is sparse or compressible,by using a small amount of signal sampling,the original signal can be reconstructed effectively by the compressed sensing reconstruction algorithm,which greatly reduces the storage space and computational complexity.With the development of compressed sensing theory,the research of image fusion algorithm based on compressed sensing theory has received more and more attention.Around the basic framework of image fusion in this paper and the careful study of relevant research results,the theory of compressive sensing and image fusion algorithms are deeply studied,the main work includes:Firstly,proposed a fusion algorithm of multi-focus image fusion based on regional characteristics.Through the single layer wavelet transform of the source image,the low frequency coefficient is not sparse,the high frequency coefficient is sparse,so the different fusion rules are adopted.Low-frequency coefficients fusion method using the regional variance of weighted and maximum absolute value;the high-frequency coefficients by random measurement matrix has better Restricted Isometry Property compression sampling.The observed value based on energy matching degree of different additive or weighted,finally the restoration algorithm is fused with the image.Experimental results show that the proposed method is simple and comprehensive to extract the characteristic information of the high frequency part,and reduces the computational complexity.Secondly,proposed a compressed sensing image fusion algorithm based non-subsample shearlet transform for Pulse-coupled Neural Network.Using non-subsample shearlet transform's multi resolution,multi-direction and shift invariant,and the Pulse-coupled Neural Network model can fully retain the details of the image characteristics.The low frequency subband coefficients using regional variance weighted matching with the combination as a fusion algorithm.High frequency in different directions and different scales by compressed sensing sampling,the observed values obtained are used as the input of the improved Pulse-coupled Neural Network;The standard deviation of the linking strength;The ignition map is obtained as the fusion operator.The high frequency subband coefficients are reconstructed by total variation method.Finally,the final fusion image is obtained by the inverse non-subsample shearlet transform.Through the evaluation results and quality analysis indicated show that the algorithm is effective.