A Hybrid Image Denoising Method Based on Discrete Wavelet Transformation with Pre-Gaussian Filtering

Abstract

Background/Objectives: At the time of acquisition and transmission noise is embedded with the images. It introduces new but unwanted information (noise) in images. The elimination of noise to analyze such data is an essential step in preprocessing. The purpose of this study is to propose a novel image denoising approach to recover original images at high noise densities without introducing unwanted artifacts. Methods: A new hybrid method based on approximation subband thresholding with pre-Gaussian filtering is presented in this study. Google Colab as a platform and python as a programming language is used for the implementation of the proposed technique. To evaluate the performance Peak Signal to Noise Ratio (PSNR) is chosen. The standard jpeg images (Cameraman, Lena, Astronaut, Cat) have been taken as an input and random noise with different noise ratios (s =0.05,0.20,0.30,0.50) is applied to get the noisy images for the experiment. In random noise scenarios, the proposed method experimented on different grayscale standard images, and performance is compared with different existing methods. Findings: The standard images with different noise ratios are denoised by the proposed method, and the quality of images is calculated in terms of PSNR. The results obtained from the proposed method on different standard images improve PSNR (PSNR= 25.80dB, s =0.50) at high noise levels significantly. Novelty: Gaussian filter improve the quality of images. However, when wavelet decomposition is blended with filtered image and thresholding is applied on approximation band improved the quality of images. Hence, the proposed method has a wide area of application to improve image quality in the field of character recognition, agriculture, medical science, and remote sensing. Keywords: Gaussian Filter; Discrete Wavelet Thresholding; Image denoising; Image Processing