Image Denoising Basing On Wavelet Transform

The real-life images are always corrupted by noise. When the noise is strong, it can affect image segmentation, recognition and comprehension. Traditional denoising methods can filter noise, but at the same time they make the image details fuzzy. Wavelet transform has the characteristic of "mathematics microscope", thus it can not only wipe off noise but also retain the image details. In many kinds of wavelet denoising methods, Donoho's wavelet shrinkage is widely used for image denoising, but this method tends to kill too many wavelet coefficients that might contain useful image information, and the reconstruction error is a little bigger. Therefore, aiming at the threshold selection, people make much research.Aiming at the defect of Donoho's VisuShrink, we improve Lakhwinder Kaur's NormalShrink and apply the improved threshold value to the dyadic discrete wavelet transform for image denoising. The experimental results show that the adaptive wavelet threshold denoising method outperforms Donoho's VisuShrink and Lakhwinder Kaur's NormalShrink in SNR and vision effect.When the high frequency information (fine edge or texture) of the signal is much, the double-frequency characteristic of the wavelet transform makes the high frequency part not well decomposed and expressed. However, the wavelet packet transform can describe the signal better because it can make fine segmentation on high frequency part. This paper adopts an adaptive threshold value denoising method based on wavelet packet transform, which gets a better result than the discrete orthogonal wavelet transform threshold denoising method in performance parameters and vision quality.The image signals deposed of by the orthogonal wavelet transform threshold can appear Gibbs impact, whereas, the reconstruct after the stationary wavelet transform adopts even sample and odd sample basing on wavelet coefficientsn and averages them. The method applied in image threshold denoising can well depress the Gibbs impact and make the denoised image preserve better edge character and better visual effect, and increase the signal-to-noise. The experimental results also demonstrate the method.