Denoising of Thermal Images Using Deep Neural Network

Abstract

In the recent scenario, there are various methods for imaging the human body for medical diagnostics, viz. X-ray radiology, magnetic resonance imaging (MRI), ultrasound, computerized axial tomography (CAT), positron emission tomography (PET) scanning, etc. Imaging the human body using IR is non-invasive as it is not ionizing in the tissue, nevertheless, it suffers from the limitation of less penetration power in the tissue and poor resolution. Thermal imaging is one of the imaging techniques, which has shown a promising result in the diagnosis of cancer. The use of thermal radiations for cancer diagnosis is safe as compared to many other imaging modalities. With the advancements in image processing routines, it has become feasible to make use of non-invasive nature of thermal radiations for cancer diagnosis. The time for the capturing of thermal image, pre-processing, and analysis of image data has improved from last few decades with recent advancements in sensor technology and analysis tools. Thus, it has become an important task to remove noise from the thermal image and restore a high-quality image in order for better thermographic assessment of images. This paper presents the denoising of thermal images using deep neural network by adding the different types of the noises to the original thermal image. The quantitative analysis is done using the three metrices which are peak signal to noise ratio, structural similarity index measurement, and mean square error in which the deep neural network shows a promising result and remove a lot of Gaussian noise and also improve the image quality than normal filtering techniques.