Image De-Noising and Compression Using Statistical Based Thresholding in 2D Wavelet Transform

Abstract

Images are very good information carriers but they depart from their original condition during transmission. They are corrupted by different kinds of noises during communication and transmission. All the images need to be de-noised before processing. The aim is to de-noise an image such that minimum amount of information is lost and maximum amount of noise is reduced. For de-noising different kind of techniques is applied i.e. linear, non-linear, adaptive and non-adaptive techniques are observed. Soft thresholding, hard thresholding, universal thresholding (Visu shrink) are used but the performance is not still good. De-noising through two dimensional discrete wavelet transform is three step process: wavelet decomposition, wavelet thresholding and wavelet reconstruction. The discrete wavelet transform gives the sparse representation of the image which is very best for the optimal threshold value selection. We used statistical based thresholding methods for de-noising which shows improved results than existing techniques. The goal of this research is to compare the performance of different statistical based thresholding techniques. The results of using these wavelet bases are compared on the basis of peak signal to noise ratio and mean square error. The research shows that use of bi orthogonal wavelets bases is better than orthogonal wavelet bases. We used bi-orthogonal wavelets version 6.8 improved the results by statistical thresholding methods.