Elimination Of Remote Sensing Image Noise Based On Wavelet Transform

The wavelet transform has emerged as an exciting new tool for statistical signal and image processing, e.g. image denoising. It has several attractive properties: multiresolution, time-frequency localized analysis and so on. These properties are of great benefit to denoising for remote sensing image. Based on the dimensions and directions of space, singular points of the coefficients and Markov Random Field in wavelet domain, three denoising methods for remote sensing image are proposed in this thesis. Experimental results show that these denoising methods are effective both in reserving the edge and in removing noise. The dissertation includes the following categories:(1). The denoising method, which proposed on multi-dimensions and different directions processing in wavelet domain includes four directions ID and a 2D wavelet transform denoising. The threshold is optimized in this chapter.(2). A new denoising method, which distinguishs the coefficients' extremums that belong to image and the coefficients' extremums that belong to noise by the tracking matrixes of the extremums in wavelet domain is proposed. The coefficients' extremums that belong to image have transmission property from coarse to fine scale, but the coefficients' extremums that belong to noise have not. We evaluate The tracking matrixes of the extremums different number base on the transmission property of each extremums from coarse tofine scale. The tracking matrixes of the extremums express the different transmission property. By the tracking matrixes of the extremums , we differentiate the wavelet coefficients' extremums and then remove the coefficients' extremums which belong to noise. Experimental results show that the denoising method is effective both in reserving the edge and in removing noise.(3). In wavelet domain we translate intrascale dependencies (contextual restriction) of coefficients into a labeling question. Based on game theory, coefficients labeling algorithm on league game is proposed. The algorithm can produce good optimization results. It also saves greatly computing time.(4). On clustering and persistence properties of wavelet transform, we research interscale and intrascale dependencies of coefficients. Under the knowledge of game theory, numeral Statistic and Markov Random Field, we correct the coefficients of wavelet transform and restore that of real image in wavelet domain. Improve image fidelity by this method.However, there are still some aspects such as wavelet function choice, distinguish the wavelet coefficients of the image from that of noise in wavelet domain, parameter estimation of the coefficients' probability density function, initialization of coefficients labeling, search method, which merit furtherinvestigation.