Remote Sensing Image Super-resolution Study Based On PDE Model

Single remote sensing image super-resolution is currently a base area of research,and is widely applicable in many fields.In this paper,a adaptive based on PDE model and sparse representation based on PDE model image zooming method for single remote sensing image super-resolution are resented.The goal is to acquire the hige resolution images of sharp-edged.The main contributions are as follows:1.The proposed adaptive mixture diffusion automatically synthesizes the linear diffusion of a Tikhonov penalty term and the improved self-snake model by means of the weight function for edge gradients.Weight function satisfy greater than 0 and less than 1,the model degenerates into the Tikhonov penalty term when the function tends to 0 and becomes the improved self-snake model when it is close to 1.The effectiveness of our model is demostrated for both Tikhonov penalty term,which tends to produce blurring in texture regions of an image,and self-snake model,which is a nonlinear level-set curve evolution approach based on active contour curvature,is sensitive to the initial location of the active contour and cannot converge in regions of concavity.Experimental results show that the proposed model synthesizes the advantages of the Tikhonov penalty and improved self-snake models,and obtains better PSNR and SSIM than traditional methods.The algorithm also requires less computation time than learning-based methods for remote sensing image magnification.2.Image super-resolution model based on sparse representation was proposed.We presented a new method for sparse representation,which combines the norm and probability of pixel density for punishment.The new method provides more exact value for image patch's sparse representation.Then we proposed the new PDE model for post-processing.It embraces three part,the first and second part ensure the result closing to real value.Edge of image was got well by giving first and second-order derivative at the last part.Experimental results show that the proposed model based on sparse representation is more efficient,and obtains better PSNR and SSIM than traditional methods.