Development of video quality framework for objective assessment of video sequences

Abstract

The visual information processing during acquisition, storage, encoding, transmission and display may lead to quality degradation. Thus, a quality assessment mechanism is desired which can measure and control the quality of video contents. The human subjects are the ultimate receiver of video services and their quality opinions about the presented video sequences represent the true assessment of video quality. However, subjective quality assessment is a time consuming and costly process and cannot be applied in real-time applications. Thus, objective assessment of video quality as accurate as subjective quality assessment is highly desired in various types of video services. Due to the significance of quality assessment, this thesis is focused to work on video quality assessment, especially, to provide a solution for assessing the video quality blindly i.e. without using the source information. In this direction of research, we have evaluated the impact of visual structural information perception characteristic of the human vision system on the quality of video sequences and developed an accurate and computational easy general-purpose no-reference video quality method. We have considered the local binary patterns for extracting the structural information from video sequences and combined with artificial neural network based machine learning approach to develop the quality assessment framework. Further, a new video quality databases accounting for new coding technologies, distortions and diverse contents is created. In this direction, a new HD video quality database representing video acquisition distortion due to low-light noise condition and compression distortions due to H.264, H.265 and Google s VP9 codecs is created. newline