Developing A Blind Image Quality Assessment (BIQA) Model Based On Image Local Contrast Features

M. Zain, Wan Zafirah (2018) Developing A Blind Image Quality Assessment (BIQA) Model Based On Image Local Contrast Features. Project Report. UTeM, Melaka, Malaysia. (Submitted)

Text (24 Pages) Developing A Blind Image Quality Assessment (BIQA) Model Based On Image Local Contrast Features.pdf - Submitted Version Download (947kB)

Abstract

This project focuses on image quality assessment (IQA) especially when we have problems on how to assess the quality of an image without presence any of reference information. Blind IQA (BIQA) aims to appraise the perceptual quality of a distorted image without information regarding its reference image. In the past, BIQA models usually predict the image quality by utilizing the transform-based quality predictive features. This approach, however, can be computationally expensive due to the need of image transformation process. This project attempts to alleviate this by developing a transform-free BIQA model that operates based on statistical characteristics of two image local contrast operators namely Gradient Magnitude (GM) and Laplacian of Gaussian (LOG). Relevant quality predictive features were first extracted based on image local contrast operators statistical characteristics. A quality prediction model was then developed through support vector regressor (SVR) utilising the extracted features. The model's performance was analysed through comparison with several available BIQA models in terms of prediction accuracy, generalisation capability as well as computational requirements.

Actions (login required)

View Item