Blind image quality assessement model based on multi-task learning

Ab Khalid, Norzawani (2022) Blind image quality assessement model based on multi-task learning. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Blind image quality assessement model based on multi-task learning.pdf - Submitted Version Download (327kB)
	Text (Full text) Blind image quality assessement model based on multi-task learning.pdf - Submitted Version Restricted to Repository staff only Download (1MB)

Abstract

The project focuses on an image quality assessment (IQA) model that estimates the quality of an image without the presence of reference information. Current blind IQA (BIQA) models typically learn their prediction separately for different image distortions, ignoring the relationship between the learning tasks. As a result, a BIQA model may have great prediction performance for images affected by one particular type of distortion but is less effective when tested on others. This project aims to address this limitation by training a new BIQA model simultaneously under different distortion conditions using a multi-task learning (MTL) technique. Given a set of training images, the model will first extract spatial domain BIQA features. The features will then use as an input to an MTL framework to simultaneously learn prediction models for different distortion classes. The predicted quality scores from each distortion class are to be weighted by the probability estimates of each distortion present in the image to yield the overall image quality score.

Actions (login required)

View Item