Document Type
Conference Item
Publication Date
1-1-2017
Abstract
This study presents a Convolutional Neural Network (CNN) model to effectively recognize the presence of Gaussian noise and its level in images. The existing denoising approaches are mostly based on an assumption that the images to be processed are corrupted with noises. This work, on the other hand, aims to intelligently evaluate if an image is corrupted, and to which level it is degraded, before applying denoising algorithms. We used 12000 and 3000 standard test images for training and testing purposes, respectively. Different noise levels are introduced to these images. The overall accuracy of 74.7% in classifying 10 classes of noise levels are obtained. Our experiments and results have proven that this model is capable of performing Gaussian noise detection and its noise level classification.
Keywords
Image noise, Noise detection, Convolutional neural networks, Training Gaussian noise
Divisions
fac_eng
Funders
Fundamental Research Grant Scheme (FRGS) grant from Malaysian Ministry of Higher Education [FRGS/1/2016/ICT03/UM/02/1]
Event Title
2017 IEEE Region 10 Conference (TENCON)
Event Location
Penang, Malaysia
Event Dates
5-8 November 2017
Event Type
conference
Additional Information
Conference paper