@ENOISIMG OF NATURAL IMAGES

Abstract

Some nonlinear, morphological and wavelet transform based denoising algorithms have been implemented in this dissertation to suppress additive white Gaussian noise (AWGN) and impulse noise, also called salt and pepper noise (SPN) effectively. In literature, many efficient digital image filtering techniques are found that perform well under low noise conditions. But their performance is not so good under moderate and high noise conditions. Thus, it is felt that there is sufficient scope to investigate and develop quite efficient but simple algorithms to suppress the noise with high density also. When an analog image signal is transmitted, it gets corrupted with AWGN in the channel. After thresholding at the receiver, the signal gets corrupted with SPN as well. Taking this problem with communication systems, e.g. television, photo phone, etc., it is very important to realize that excellent denoising algorithms are required to suppress both type of noise quite effectively without distorting the edges and fine details of the image. The problem is taken up and efforts are made to develop quite efficient noise removal techniques. Denoising is usually required to be performed before display or further processing like segmentation, feature extraction, object recognition, texture analysis, etc. This dissertation contains mainly two parts. Impulse noise is considered in the first part of dissertation in which impulse noise removal techniques based on mathematical morphology and some decision based are mentioned as well as simulated with the four standard images of size (512x512) such as Lena, Boat, Cameraman and Flinstones images and compared their performance on the basis of performance indices like peak signal to noise ratio (PSNR), mean squared error (MSE) and image quality index (IQI). In the second part of the dissertation additive Gaussian noise is considered in which Gaussian noise reduction techniques based on wavelet transform have been examined and simulated with all above four images and performance of these algorithms is measured in terms of various performance evaluation parameters. So wavelet based neighborhood thresholding (NT) and window thresholding (WT) are suggested to suppress moderate and high additive Gaussian noise from an image. In this technique the threshold is applied on a particular window of the images not whole the images after taking the wavelet transform. For suppressing the impulse noise morphological based algorithm and PDB algorithm are proposed to get the better result. At the end, observation on comparative study of all these techniques has been presented.