A Study on Various Common Denoising Methods on Chest X-ray Images

Abstract

Image 'denoising' is an important pre-processing step as noise affects the image quality. The behaviors of denoising algorithms have been examined on a set of 2D Chest X-ray (CXR) images. Both the 'Linear' and 'non-Linear' denoising methods are considered, each having two techniques, such as (i) Moving average and (ii) Gaussian under the Linear method, and (iii) Median and (iv) Bilateral under the non-Linear method, respectively. Noise variance and Signal-to-Noise Ratio (SNR) are computed on the raw images. The study finds that, Median filter is the best of the lot with the PSNR mean = 34.5213; PSNR σ = 0.9618; MSE mean = 12.2337; MSE σ = 1.5491; and Big(O) = 0.02873 milliseconds. Finally, a comparative study has been made among these denoising methods to predict five hundred Tubercular CXRs, which shows that, using the Median filter, 92% CXRs can be accurately diagnosed, followed by Gaussian, Bilateral, and Moving average type filters with the respective accuracies of 80%, 72%, and 60%, respectively. This work could be a ready-reckoner to the researchers in choosing the best filtering technique when working on Medical X-ray images.

Lightweight Malicious Code Classification Method Based on Improved SqueezeNet

“…Deep neural networks already have more mature applications in many areas, for example, in medicine [7], robots [8], games [9], stock prediction [10] and image processing [11,12], etc. Without human intervention, neural networks can automatically extract features.…”