Trace-norm regularisation-based learning framework for blind image quality assessment model

Jahirin, Rachel Lyn (2024) Trace-norm regularisation-based learning framework for blind image quality assessment model. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full text) Trace-norm regularisation-based learning framework for blind image quality assessment model.pdf - Submitted Version Download (1MB)

Abstract

The project’s focus is on developing an image quality assessment (IQA) model that can accurately estimate an image’s quality without the need of a reference image. The current blind IQA (BIQA) model typically trains their prediction separately for different image distortions, without considering the relationship between these learning tasks. Consequently, a BIQA model may perform well when tested on images affected by one type of distortion, but it may not be as effective when tested on other distortions. This project aims to overcome this limitation by simultaneously training a new BIQA model under different distortion conditions using the trace-norm regularisation-based learning framework. The model first extracts spatial domain BIQA features from a set of training images, and these features are then used as inputs to the trace-norm regularisation-based learning framework to learn prediction models for different distortion classes. The model then combines the predicted quality scores from each distortion present in the image to yield the overall image quality score.

Actions (login required)

View Item