Deep neural network-based noise removal

Liew, Ting Foo (2023) Deep neural network-based noise removal. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Deep neural network-based noise removal.pdf - Submitted Version Download (2MB)

Abstract

This thesis covers the design of Deep Neural Network based noises removal by using Convolutional Neural Network architecture. The goal for this project is to validate the performance of peak signal noise-ratio before and after the process of the image, and remove Gaussian noise from noisy image, and output an improved image quality with edge detail preservation as much as possible, Using CNN as a trained model, design a Matlab code to perform denoising by adjusting the parameter of the training option and dnds datastore. Some of the traditional method such as Block Matching & 3D filtering, BM3D and Expected Patch log-likelihood method, EPLL is worse compared to CNN, these methods also have their own problems which is low image quality, low Peak Noise Signal-Ratio, etc. That is why Convolutional Neural Network related methodologies is created to enhance the denoising performance. This network architecture utilized the technique of deep learning to denoising the image. So, this thesis is to review CNNs model for image denoising approaches and resulting in better image quality with high PNSR value.

Actions (login required)

View Item