The Analysis Of Statistical Characteristics Of Remote Sensing Image And Its Application In Image Restoration

Because of abundant content and wide format,remote sensing image denoising faces many challenges.Using the statistical characteristics of the image block is an important way to improve the performance of image denoising.There are two kinds of the statistical characteristics:interior characteristics from similar block of the being-processed image and exterior characteristics from external similar image block.Different statistical characteristics have the advantages and disadvantages on the image denoising.To make the best use of their respective advantage,this dissertation is aimed at combining interior characteristics and exterior characteristics for remote sensing image denoising.This paper has studied on whether denoising in transform domain can get better performance,and proposed a denoising algorithm that combines the interior and exterior characteristics in PCA transform domain.This algorithm uses interior and exterior characteristics based on a denoising method for natural image in PCA transform domain.Previous studies and preliminary experimental results show that,the combination of interior and exterior characteristics uses enables to denoise for remote sensing image but also needs further research.Secondly,we have studied how to select either the interior or the exterior characteristics according to the image content.And we proposed a new method to remote sensing image denoising by training an adaptive combination model.According to the number of similar blocks,it chooses denoising method using interior or exterior characteristics adaptively.Experimental results show that the denoising effect for remote sensing image of this algorithm is better than the methods using a single kind of statistical characteristics and combination method based on Patch-SNR.At last,in order to use the information better,this paper has proposed a remote sensing image denoising method based on RTF model for combining interior and exterior characteristics.The new algorithm fuses respective results using multiple characteristics,and regresses for the best denoising result.Because of the effective use of the mutil-cell and multi-hierarchical information associated with the being-processed image,the performance of the denoising algorithm is not only better than those methods using a single kind of statistical characteristics,but also better than the method by training an adaptive combination model.