OPTIMIZATION OF THE SPARSE REPRESENTATION PARAMETERS FOR THE FUSION OF REMOTELY SENSED SATELLITE IMAGES

Abstract

Abstract. Image fusion methods are widely used in remote sensing applications to obtain more information about the features in the study area. One of the recent satellite image fusion techniques that can deal with noise and reduce computational cost and deal with geometric misregistration is sparse representation model. The important part of creating a generalized sparse representation model for satellite image fusion problems is defining initial constraints and adjusting the corresponding regularization coefficients. Regularization coefficients play an essential role in the performance of the sparse representation model and convergence of the optimization solution. Also, the number and size of sub-images extracted from the dictionary matrix in the sparse representation model, and the number of iterations of the optimization step are important in building a sparse representation model. Therefore, in this research, the four parameters that affect the performance of the sparse representation model were investigated: the number of sub-images, the size of sub-images, regularization coefficients, and the number of iterations. Results obtained from pan-sharpening of OLI-8 images showed that optimal values for the number and size of sub-images, regularization coefficients, and the number of iterations were equal to 150, 9×9 pixels, 10-4, and 4 respectively. Results from this study can be generalized to other satellite image fusion problems using sparse representation models.