Skip to content

KUCognitiveInformaticsLab/zhu_project_real_and_fake

Repository files navigation

Real and Fake Project

Master Thesis Thema:

  • Comparison of the Detection Sensitivities of Human and Full-Reference Image Quality Assessment Models to Distortions in Natural and Non-natural Images

Measure JND by Psychopy

  • We decide to use 2AFC (Two-Alternative Forced Choice) QUEST procedure (a bayesian adaptive psychometric method) Find the Just noticeable difference (JND)- minimum detectable distortion intensity

Heeger-Bergen texture synthesis algorithm

  • Retain the first-order statistics of both its pixel intensity and responses to multi-orientation and multi-scale filters

Models

  • Traditional:
  • MSE, NLPD : low-level features
  • SSIM, MS SSIM: structure information
  • DL-based:
  • LPIPS: deep features extracted by VGG-16
  • DISTS: the combination of deep features and structure information

About

The issues when we use the PsychoPy software

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published