Simulation and Analysis of Image Denoising Techniques

Abstract

In the present era, visual information transmitted in the form of digital images is becoming a major method of communication, but the image obtained after transmis- sion is often corrupted with noise. The received image needs preprocessing before it can be used in applications like segmentation, feature extraction, object recognition, texture analysis etc. Image denoising involves the manipulation of the image data to produce a visually high quality image. Despite the signi cant research conducted on this topic, the development of e cient denoising methods is still a compelling challenge. In this research, various standard image denoising algorithms are imple- mented and two new algorithms are proposed. Since medical and synthetic aperture radar images are corrupted by speckle noise while natural images are corrupted by gaussian and random noise, di erent noise models like speckle, gaussian and random are considered for experiential analysis. In this research work, two image denoising techniques like Novel statistical hybrid lter in spatial domain and modi ed curvelet in transform domain are proposed and they are compared with standard image de- noising techniques. Novel statistical hybrid lter which is hybridization of mean and median lter, is used for removing speckle noise from grayscale images like ultrasound and synthetic aperture radar. Proposed curvelet is used for removing speckle, gaus- sian and random noise for both grayscale and color images. Simulation results show that proposed methods have better visual quality as compared to earlier reported methods available in literature. This result also demonstrated using objective evalua- tion parameters like PSNR, COC, RMSE, TE. The test images used for experimental analysis are Ultrasound, Synthetic aperture radar, Lena and Peppers. As per results, we conclude that proposed novel statistical hybrid lter is performing better for gray values 2N 2N images in PSNR, RMSE, COC and execution time point of view. The simulation results show that the proposed curvelet method outperforms the various wavelet thresholding methods for denoising of both grayscale and color images.