Research On Compressive Sensing Based Image Fusion

Image fusion is an image processing technology that combines multiple images from the same scene, aiming at integrating complementary information of multiple images to obtain better image description of the scene. Nowadays, the compressive sensing theory provides a new thought for signal processing method. According to compressive sensing theory, compressible signal or image is available for a particular measurement matrix which projects the signal from high dimensional space to low dimensional space, which reduces dimension of the signal or image. Then the original signal or image is recovered by solving optimization problem. In the field of image fusion, studies of fusion algorithm and its related application have made great progress, but information fusion based on compressive sensing theory also needs to be further improved. This paper mainly studied the multi-focus image fusion for single pixel camera, image fusion based on approximate sparse representation and remote sensing image fusion with image super-resolution based on sparse representation. The main contents are listed as follows:1. A novel multi-focus image fusion method based on single pixel camera is proposed. Because of random projection cannot include the geometry of original image, we cannot directly measure the compressing imaging measurements of salient information. Therefore, in proposed algorithm, firstly, several bases from Hadamard matrix is selected as clarity measurement bases by using Wilcoxon rank test method. Then,clarity measure coefficient is obtained by compressed sensing theory,which guides compressive measures to be fused. Finally, fused image is gotten by reconstructing fused compressive measures.2. An image fusion method based on approximate sparse representation is proposed. In the application, the existing image fusion methods based on sparse representation are more time-consuming and affect the running efficiency. With proposed algorithm, first of all, source images are segmented into some image blocks by a sliding window. Then,the coefficients obtained by using the approximate sparse representation algorithm to solve different source image blocks can guide the corresponding image block to fuse. Finally, resulting image is obtained by refactoring fusion image blocks. This method not only improves the fusion quality, but also greatly reduces the running time.3. A remote sensing image fusion method via image super-resolution based on sparse representation is proposed. The traditional remote sensing image fusion methods don't make full use of the spatial information of low-resolution multispectral image. In order to solve this problem, this paper gives full consideration to spatial information of multispectral image. First, the low-resolution multispectral image is processed by image super-resolution based on sparse representation. Then,the enhanced brightness component of the multispectral image and panchromatic image is fused by stationary wavelet transform. Finally,fused image is obtained by YUV inverse transform. Experimental results show the proposed algorithm can effectively improve spatial details of fused image and preserve spectral information at the same time.