Full Reference Objective Video Quality Assessment with Temporal Consideration

• Main Author: Loh, Woei Tan
• Format: Thesis
• Language: English; English
• Published: Universiti Malaysia Sarawak (UNIMAS) 2018
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://ir.unimas.my/id/eprint/26581/1/Full%20Reference%20Objective%20Video%2024pgs.pdf; http://ir.unimas.my/id/eprint/26581/4/Loh%20Woei.pdf; http://ir.unimas.my/id/eprint/26581/
• Institution: Universiti Malaysia Sarawak

Video quality assessment (VQA) is an extension of image quality assessment (IQA). A video is a series of images arranged in time sequence. Therefore, IQA methods can be used to assess videos quality. A video has three dimensional data; two for the spatial dimensions and one for the temporal dimension. IQA methods assess video quality by assessing spatial effects without the need to consider the temporal effects and distortions. This makes IQA methods inappropriate and maybe inaccurate for assessing video quality. In order to apply in real time scenarios, VQA methods have to be reliable and correlated well to the judgement of human visual system (HVS). Furthermore, they have to be computationally efficient to give fast results. Current VQA methods have good correlations with subjective scores but are high in terms of computational complexity. In this thesis, two VQA methods, Index1 and Index2, with lower computational complexity are proposed. Index1 deals with Just Noticeable Difference (JND) in both spatial and temporal parts of the video. For the temporal part, JND is combined with temporal information to account for temporal distortions. For Index2, it is based on the previous work of mean difference structural similarity index (MD-SSIM). The temporal part of Index2 deals with the variation of temporal information. Both of the proposed methods are then compared with state-of-theart VQA methods in terms of performance and computational complexity. The proposed methods were found to have acceptable performance with lower computational complexity.