The Research Of Remote Sensing Image Denoising Method

The digital image in the process access and transmission is generates the noise image byimaging equipment and external environment noise interference effect, reduce digital imagenoise process is called image denoising. The Nonsubsampled Contourlet Transform (NSCT)is a new tool, it is a kind of multi-resolution, multi-scale, Multi-direction of imagerepresentation method. It can accurately capture the image edge profile information of thedifferent scales and different direction with the sub-band, it has a very wide development andapplication prospect in image processing area.A new method for image denoising based on the Nonsubsampled Contourlet Transform(NSCT) and the Linear Decreasing Weight Particle Swarm Optimization (Lin-WPSO) isproposed in chapter three. This method firstly acquires the optimal threshold of the GCV riskfunction by using Lin-WPSO in the NSCT domain, and then removes the noise through softthreshold function, which doesn’t need the prior information of noise variance. The proposedmethod can improve the image of the peak value signal-to-noise ratio.A new method for image denoising based on the free distributed hypothesis testthreshold (FDR) and the nonsubsampled contourlet transform (NSCT) is proposed in chapterfour. This method firstly acquires the free distributed false discovery rate hypothesis test instatistics to set the threshold in the NSCT domain, and then removes the noise through softthreshold function, which doesn’t depend on the length of the signal. The experimental resultsshow that this method utilizes the shift invariant of NSCT transform to inhibit the pseudoGibbs distortion effect.Through two specific change detection algorithm, summarized the change detectionalgorithm for general research ideas. This is the ultimate goal of this thesis.