Video quality measurement based on 3-D. Singular value decomposition

• We proposed a new video quality measurement based on considering whole video as a 3-D tensor.
• The original video were decomposed to three orthogonal matrices used Tucker algorithm as its basis.
• The distorted video was reflected into original video basis.
• We proposed an equation to weight the difference between the original and reflected singular values.
• The experimental results showed a high correlation between subjective tests and the proposed algorithm results.

This paper presents a new full reference Video Quality Assessment (VQA) method based on using 3 Dimensional Singular Value Decomposition (3-D SVD). The method compares the structural properties and the luminance characteristics between the reference and the distorted videos. This aim is obtained by applying 3-D SVD that is singular value decomposition in a 3-D space. In principal, the distorted and the original videos are projected on the singular vectors of the original video. The weighted difference between the reflections coefficients could be considered to quantify the quality of videos. For our experiments, we have used the LIVE and EPFL-PoliMI video quality databases to evaluate the performance of our metric. The results show a great correlation between the measure scores and the subjective scores.