The Research Image Denoising Method Based On Wavelet Transform

Digital image is usually contaminated by the noise in its acquisition and transmission,which makes it very difficult to extract information details. Therefore it is necessary tosuppress the noise in image before next processing.Wavelet analysis is a method of time-frequency analysis developed in the1980s, along withthe development of the wavelet theory. Wavelet analysis has been used in a lot of fields, andachieved great success. One of its most important applications is image processing. It hasbeen shown that wavelet transform is one of the most powerful tools in image processing.Wavelet analysis has been widely used in image denoising, compression, quality enhancementand edge detection.This essay has mainly focused on image denoising. The basic theories of wavelet transformare introduced, for example Multiresolution Analysis, Mallat algorithm. There are threefiltering methods, Modulus maxima filtering, Spatial correlation filtering, and waveletthreshold filtering for image denoising in the wavelet domain. The wavelet threshold filteringis widely used because it can be easily implemented and has low computational complexity.In this essay, we have focused on wavelet threshold filtering algorithm, whose performancedepends on how to select the threshold function and the threshold. With respect to thethreshold function, the traditional hard and soft threshold functions result in the discontinuityand permanent bias. In this essay, a medium-soft threshold function based on waveletshrinkage is presented. Simulation results have shown that the modified function improves thedenoising effect comparing with the other threshold functions. The DJ threshold is a classicthreshold in image denoising. However, if the image contains a lot of details, the DJ thresholdis usually greater than the true value, which degrades the denoising effect. Some researchershave proposed the modified methods of noise estimation based on a single image. This essayhas proposed to estimate the noise intensity from two noisy images. It has been shown that theerror is very small between the estimated value and the true value and this method isespecially superior to the DJ threshold when a lot of details are contained in images.Because the discrete wavelet transform (DWT) can generate the Pseudo-Gibbs phenomenonin image denoising, in order to improve the deniosing effect, the stationary wavelet transform(SWT) is adopted due to its time-invariance property. With the increase of decompositionlevels, the noise intensity decreases, therefore the adaptive threshold is used to furtherimprove the denoising effect. The PSNR (peak signal to noise ratio) and MSE (mean squareerror) are usually used to evaluate the image denoising quality. We have compared our improved method with the traditional ones and simulation results show that our method hasbetter performance.