Performance Analysis Of A Visual Saliency-Guided Model For Blind Image Quality Assessment (VBIQA)

Quah, Swee Liin (2018) Performance Analysis Of A Visual Saliency-Guided Model For Blind Image Quality Assessment (VBIQA). Project Report. Universiti Teknikal Malaysia Melaka, Melaka,Malaysia. (Submitted)

Text (24 Pages) Performance Analysis Of A Visual Saliency-Guided Model For Blind Image Quality Assessment (VBIQA).pdf - Submitted Version Download (3MB)

Abstract

The project focuses on image quality assessment (IQA), especially in the problems of predicting the quality of an image blindly or in the absence of the information of reference images and types of distortion present in the image.Nowadays,IQA is integrated in various range of applications such as compression,transmission, restoration,etc.Over the last decade,there are rapid growth in developing blind image quality assessment(BIQA) approaches and various models have been introduced. Previous BIQA models often extract handcrafted quality predictive features from randomly sampled image patches. The extracted features are then used to learn their quality prediction model.This approach,however,ignore a human visual system (HVS) property that can have major impact to overall perceptual image quality.The property, referred to as visual attention,shows that the viewer may be more likely to detect distortions in highly salient regions,as compared to regions of low saliency.Thus,this project aims to incorporate this property into a BIQA model by utilising an image's saliency map to guide the model sampling process instead of sampling randomly.A new BIQA model that correlated highly with human judgements of quality is developed at a level that is competitive to the previous BIQA models.

Actions (login required)

View Item