Image Denoising Method Based On Multiwavelet And Bandelet Transform

There is always noise data during the process of acquiring and transmitting image data. Only through denoising process can represent the original image data effectively, so the image denoising has become a hotspot in the research of image processing. Wavelet denoising has achieved relatively better effect due to its characteristic of low-entropy and multi-resolution. The characteristics of compact support, orthogonal, symmetry and vanish moments which scalar wavelet function can not possess simultaneously are of great importance in wavelet theory. Instead, multiwavelet shows better capability in these characteristics. In the mean time, the traditional wavelet is not good at dealing with the line characteristic of images. In order to solve this problem, researchers have come up with some new multi-scale analysis methods bases on wavelet theory, such as Bandelet.This paper introduces image denoising technique based on mutiwavelet and Bandelet Transform. We introduce and analyze multiwavelet theory and Bandelet theory in detail, state some results of threshold denoising methods which will be used in ours Multiwavelet-Bandelet denoising method. At first, Based on the property that Bandelet shows better result in representing geometric regular images, we design a Bandelet denosiding algorithm in which traditional wavelet transform is replaced by Bandelet transform; in order to enhance the accuracy of computing the geometric flows, we propose a procedure that uses wavelet-denoised image to compute the geometric flows and its dyadic square segmentations of an noised image. Because the Bandelet do not possess the translation invariant property, we introduce the translation invariant process into the Bandelet denoising algorithm so that the pseudo-Gibbs phenomenon can be suppressed effectively. Finally, a threshold denoising algorithms based on multiwavelet and second generation Bandelet Transform together with translation invariant procedure are presented in this paper, further more, we discuses the rules of choosing prefilter of the multiwavelets transform during the multiwavelet-Bandelet denoiding algorithm.The numerical experiments show that as for the geometric regular images, the Bandelet approach for image denoising is much better than wavelet method. Further more, the Multiwavelet-Bandelet approach for image denoising, which can suppress the pseudo-Gibbs phenomenon effectively and preserve much more details of the images, shows better results than Bandelet method.