| Remote sensing image contains many important information of the earth’s surface, is of great significance on the regional planning, environmental monitoring and engineering construction. In the process of image acquisition and transmission,remote sensing images exist with kinds of noises, which reduce the image quality,then influence the follow-up image interpretation, etc. Therefore, filtering processing to noises must be done to give images of high quality in order that later analysis,identification and high-level processing should be convenient.Wavelet analysis has the time-frequency localization and multiscale analysis property, thus making it possible to effectively extract information. In 1994, Donoho proposed the soft-threshold and hard-threshold de-noising methods, by setting the threshold threshold to deal with high frequency coefficient, achieve the image noise filtering. Wavelet packet inherits the advantages of wavelet analysis, can also further decomposition of the wavelet decomposed high frequency parts, so the signal feature extraction is more effective.SVM can solve the classification problem of small samples and high dimensions,and has a strong generalization capability, so it has been widely used in classification problems. In an SVM classification algorithm, fuzzy support vector machine gives different membership degree to different samples, so as to improve the classification ability of containing abnormal data problem.On the basis of the conbination of the wavelet packet de-noising and the fuzzy support vector machine theory, this paper seeks to put forward a new de-noising method. The basic steps of the new de-noising method are: firstly, we use wavelet packet threshold to deal with a noise-containing image; secondly, we use the wavelet packet decomposition on it; thirdly, the coefficients of the wavelet packet decomposition are classified into the original image part and the noise part, usingfuzzy support vector machine; finally, the wavelet packet coefficients of the image are reconstructed to recover the image from its noisy version contaminated by noises. As to how to select and set the parameters of SVM and the fuzzy membership, this new method uses the training sample cut set strategy combined with genetic algorithm method to select the parameter of kernel function and penalty factor and improve the traditional fuzzy membership degree.This thesis also analyzes marcato with effect illustrations the application of the new de-noising method to different noise types. The result of the simulations shows that the new de-noising method is more effective in preserving the edges of the image. |
