Off-campus SJSU users: To download campus access theses, please use the following link to log into our proxy server with your SJSU library user name and PIN.

Publication Date

Spring 2025

Degree Type

Thesis - Campus Access Only

Degree Name

Master of Science (MS)

Department

Applied Data Science

Advisor

Vishnu Pendyala; Mohammad Masum; Subir Varma

Abstract

This project focused on image denoising using deep Convolutional Neural Network (CNN) architecture and Large Language Models (LLMs). A study was conducted to investigate different CNN and LLM models and their effectiveness in denoising images. Large Language Models (LLMs) are mainly used for Natural Language Processing, but recent advances in LLMs have allowed some of them to accept images as input. One of the objectives of the project was to see if LLMs could be used to denoise images and compare the performance with that of CNN models. LLMs work differently from CNNs because the input is fed into an LLM as a prompt. Training and test datasets were created for this project by adding Gaussian noise samples to an existing image dataset from a public domain database called Kaggle. Existing models used for the investigation were ChatGPT, BRDNet, and ECNDNet. ChatGPT is an LLM, and it was observed that LLM models could be used for image denoising. A new model called the Enhanced Model was created, trained, tested, and compared with these models. The metrics used to measure how well each image was denoised were peak signal-to-noise ratios (PSNR) and structural similarity index measurement (SSIM). The Enhanced Model produced the highest average PSNR and SSIM values for all the images tested.

Available for download on Tuesday, August 06, 2030

Share

COinS