The Theory And Method Of Remote Sensing Image Fusion Based On Joint Sparsity Representation

Due to the limitation of optical sensor technology,the satellite can not provide high-resolution multispectral images, which is a critical technology bottleneck restraining the research in earth observation. As a software solution,remote sensing image fusion merge many images into a high-resolution and colorful synthetic images for human visual judgment and the follow-up image analysis, which comes from one or different sensors for a same scene. Therefore, one of the hot issues in the field of image fusion is how to make full use of complementary information and redundant information between the source images to reconstruct the fused image rapidly and accurately.In recent years, the sparse representation has been widely used in remote sensing image fusion and achieved some effectiveness. However, the traditional fusion method based on sparse representation usually ignores the correlation between the source images, and has high time complexity of the dictionary training. In order to solve these problems, this paper employs the structural dictionary learning and joint sparse representation to modify and improve the remote sensing image fusion method. The contents of this paper are summarized as follows:1. In order to reduce the complexity of the algorithm, this paper proposes a novel image fusion method based on structured dictionary learning. First of all, the corresponding structured dictionaries are trained from the panchromatic image by double-sparsity model.Secondly, we employs sparsity representation to extract the detail components of the panchromatic image. Finally, the fused image is constructed by injecting the detail components into the low-resolution multispectral image. Compared with the classical dictionary learning algorithm, in the process of dictionary learning, our method takes full account of the structural correlation between the dictionary atoms so that reduces the number of iterative, which improves the efficiency of training. Morever,the learned compact structural dictionary can further improve the calculation speed of the subsequent image processing.2. Aiming at the problem of local block effect distortion and the low contrast in the fused image, this paper proposes a novel image fusion method based on joint sparse representation. The traditional injection fusion methods often ignore the problem of redundant details and low frequency information from the panchromatic image. Therefore,we first employs the joint sparse model to obtain the innovation sparse coefficients of the panchromatic images; then choose the needed innovation sparse coefficients by the absolute value principle, and reconstruct the innovation components of the panchromatic image; finally the fused image is constructed by injecting the innovation components into the low-resolution multispectral image. Through the analysis of the experimental results of QuickBird and IKONOS satellite remote sensing images,we can see that the proposed method has better performance than other state-of-the-art methods in terms of the spectral information, spatial resolution and computational efficiency.