Subjective evaluation of approximate Discrete Sine Transform for the Versatile Video Coding standard

Abstract

For the past few years, a Joint Video Exploration Team (JVET) has been working to establish a new compression standard known as Versatile Video Coding (VVC). Several coding tools have been introduced. The Adaptive Multiple Transform (AMT), used as a residual transform coding, involves five discrete transforms that belong to cosine (DCT) and Sine (DST) families. However, transform computation using direct matrix multiplication requires much hardware resources and yields high run-time complexity. Particularly the DST-VII, frequently used by the transform module, relies on complex matrix multiplication. This results in a significant issue while implementing it in hardware. Therefore, DST-VII approximation was addressed as an efficient solution to reduce the operation count and the execution time, meanwhile preserve almost a similar coding performance to the exact transform. This paper deals with objective and perceptual assessment of the proposed transform using the Peak Signal to Noise Ratio (PSNR) and the Global Score Reduced reference Video quality assessment based on Human visual system $(GS_{RVH})$ metrics respectively.