Research Project

Video Quality Assessment Compatible with PSNR Using Human Visual Perception

This project develops perception-aware video quality metrics that integrate modern psychovisual models of central and peripheral human vision into objective video quality assessment.

Video Quality Assessment PSNR-Compatible Metric Human Visual System Psychovisual Models Visual AI

Key Contributions

HVS integration of human visual system characteristics into objective video quality prediction
PSNR+ development of a PSNR-compatible perceptual quality metric
Central + Peripheral modelling of visual sensitivity across central and peripheral vision

Abstract

This work analyses existing data on human perception of temporal frequencies in both central and peripheral vision. It presents a new method for measuring video quality based on modern models of human contrast sensitivity, including visibility thresholds for spatio-temporal sinusoidal variations with different spatial dimensions and temporal modulation rates on modern displays.


Unlike traditional pixel-based metrics, the proposed methodology incorporates recent knowledge of how humans perceive visual information across the field of view. This enables objective video quality prediction to better reflect subjective human experience while remaining compatible with established PSNR-based evaluation frameworks.


The research contributes to perceptual video compression, immersive media, embedded vision systems, UAV video transmission, and next-generation Visual AI systems where perceived quality, bandwidth efficiency, and computational constraints are critical.

Technical Focus

Metric Development

  • PSNR-compatible video quality assessment informed by human visual perception.
  • Spatio-temporal contrast sensitivity modelling for modern video presentation conditions.
  • Central and peripheral vision integration for improved perceptual prediction.
  • Objective quality prediction closer to subjective human ratings.

Application Context

  • Perceptual video compression for reducing visually redundant information.
  • Immersive and wide-angle video where peripheral perception becomes important.
  • Embedded video systems operating under bandwidth and computational constraints.
  • Visual AI workflows that benefit from perception-aware image and video processing.

Publications and Materials

IEEE IVCNZ 2024

Mozhaeva A, Vlasuyk I, Potashnikov A, Mazin V and Streeter L (2024). Video Quality Metric Compatible with PSNR Considering Recent Knowledge of Peripheral Characteristics of Human Vision. 39th International Conference on Image and Vision Computing New Zealand, Christchurch, New Zealand.

View IEEE Paper

Springer LNCS

Mozhaeva A, Mazin V, Cree MJ, Streeter L (2022). Video quality assessment considering the features of the human visual system. Lecture Notes in Computer Science, Springer Nature Switzerland, pp. 288–300.

View Springer Chapter

FRUCT 2021

Mozhaeva A, Streeter L, Vlasuyk I, Potashnikov A (2021). Full reference video quality assessment metric on base human visual system consistent with PSNR. 28th Conference of Open Innovations Association FRUCT.

View IEEE Paper

Code and Data

Source code and experimental data for the PSNR-M Plus model are available in the project repository.

View GitHub Repository

Visual Methodology

Model formula and PSNR-compatible video quality metric structure
Core PSNR-compatible perceptual quality model structure.
Framework of the methodology for weight estimation
Framework of the methodology for estimating perceptual weights in the proposed video quality metric.
Spatio-temporal component in central vision
Spatio-temporal component in central vision used for perceptual quality prediction.
Central and peripheral vision video quality assessment process
Video quality assessment process incorporating central and peripheral human vision characteristics.

Experimental Results

The proposed metric was evaluated using video sequences from LIVE-NFLX and related test scenarios, including frames with motion in peripheral areas. The results compare PSNR, PSNRper, FovVideoVDP and the proposed perception-aware metric.

PLoS related contrast sensitivity result
Contrast sensitivity model data supporting perceptual quality estimation.
Video quality metric experimental result
Experimental visual quality assessment result for selected video sequences.
PSNR model formula result
PSNR-compatible model component used in the proposed methodology.
Peripheral visual model image
Peripheral vision model component for perceptual video quality assessment.
PSNR peripheral component result
Peripheral PSNR-related component for frames with peripheral motion.
Video quality visual result example
Example visual result from the video quality metric evaluation.
Correlation interval of video quality metrics
Correlation interval of video quality metrics on LIVE-NFLX video sequences, including PSNR, PSNRper and FovVideoVDP comparisons.

Contact

Primary contact: amozhaeva@eit.ac.nz.