Reaserch On Image Denoising Based On Median Filter And Wavelet Transform

When the impulse noise accounts for a larger proportion of pixels in neighborhood, the median filtering algorithm for image denoising will be significant variation, the reason is that the noise pixel may occupy the neighborhood median position. In order to improve the filtering effect an improved median filtering algorithm is presented, before the median operations on neighborhood, noise pixel identifications in the neighborhood should be done, and the impulse interferences in the neighborhood are eliminated, the median filtering operations are used forthe rest pixels. The neighborhood variance is employed for neighborhood noise pixel judgment, but sometimes the noise variance calculation is not ideal bocause of fewer pixels, therefore the identification of noise pixels should be done in a larger neighborhood, while a smaller neighborhood is used for the median filtering operation without noise pixels, so the reliability of noise pixel identifications and clarity of the image filtering are ensured.While wavelet transform is used for eliminating Gaussian noise, the wavelet threshold filtering is an important algorithm which mainly operates for wavelet coefficients which represent the high frequency part. When the noise is larger, the filtering effect needs to be improved, this is because the low frequency image representing image brightness information division contains certain interference noise, but wavelet transform can not be done for too many levels, so this paper adopts a method which wavelet threshold filtering applied for the high frequency part of wavelet decompositon image and a averaging filtering for low frequency part. The simulation result shows that the improved algorithm has good results.For the image containing impluse and Gaussian noises at the sametime, the mixed filtering that contains two kinds of filtering method mentioned above can be used for elimating the noise.