Investigation of multiresolution image denoising schemes using wavelet transforms

Abstract

Image denoising is a fundamental problem in image processing. The main determination is to quash noise from the degraded image while keeping other details of the image unchanged. In recent years, many multiresolution based approaches have attained great success in image denoising In a nut shell, the wavelet transform provide an optimal representation of a noisy image, with information bearing signal from a small number of coefficients and noise by all other remaining coefficients. The noisy coefficients can be eliminated by thresholding procedure. Thus, every method of image denoising using wavelets has three basic steps: to compute the wavelet transforms of the noisy image, threshold the wavelet coefficients and finally to compute the inverse wavelet transform. The effectiveness of the noisy coefficient separation of the image depends on the capability of sparse representation of the image. The wavelet representation is optimally sparse, since the wavelets overlapping a singularity have a large wavelet coefficient and all other coefficients are small. Utmost the noisy wavelet coefficient shrinkage is better, only if the threshold value is properly selected. The denoised images obtained using Discrete Wavelet Transform (DWT) suffers from shift invariance and poor directional selectivity. These drawbacks can be overcome by the use of 2D-Dual Tree Discrete Wavelet Transform (DTDWT). The 2D-DTDWT converts an N-point signal into M coefficients with . newline