New method of noise removal in images using curvelet transform

The term Curvelet transform in the field of Image Processing is quite well known from past few years. Its ability to detect curved features and smooth areas in an image marks its huge importance in the area of image denoising. However the ability to denoise image depends upon the selection and application of threshold after doing Curvelet based decomposition of an image. In this paper we are presenting our research methodology based on Curvelet transform image denoising. Our approach is based on the implementation of a modified window neighborhood processing that adapt itself based on the variance of neighboring pixels. We describe the problem we are considering for our research, present a brief overview of relative literature, describe the proposed methodology we have implemented and illustrate our future plan.

I. Introduction

The term image denoising is of immense importance in the field of image processing, the motivation and notion behind image denoising is to remove the noise content from the noisy image as much as possible thereby retaining the important image features. However Wavelet transform can be an alternative for the same but since natural images contain various curved features and moreover the smoothness along the curves is also distributed quite typically. In case of 1-D or 2-D images wavelets can achieve reasonable results but for higher dimensional images it proves out to be obviously insufficient to detect the smoothness along the edges and can capture only limited information features concerned with directionality. As we know mostly edges contain more typical information content, Wavelet can signifies the edge crossing characteristics but cannot represents the characteristics of edge crossings. To overcome the inefficiency a new transform called Curvelet transform was given by Candes and Donoho which is a multiscale geometric transform that allows to represent important image features as well as directionality features simultaneously. Unlike traditional Wavelets, edges and singularity along curves Curvelet transform not only incorporates the time-frequency analysis capability of wavelets transform but also adds up the capability of identifying directionality.