3D image quality estimation (ANN) based on depth/disparity and 2D metrics

Immersive image/video services will be soon available to the mass market due to the technological advancement of 3D video technologies, which include 3D-Ready TV monitors at affordable prices. However, in order to provide demanding customers with a better service over resource limited (e.g., bandwidth) and unreliable communication channels, system parameters need to be changed “on the fly”. Measured 3D video quality can be used as feedback information to fine tune the system parameters. The main aim of this paper is to analyze and present impact of objective image quality assessment metrics on perception of 3D image/video. Neural Network statistical estimator was used to examine the correlation between objective measures on input image base and Differential Mean Opinion Score (DMOS) of used image base. For this purpose part of LIVE 3D Image Quality Database [7] was used. The results suggest that comparison of the neural network DMOS estimators based on full-reference and no-reference objective metrics shown very similar behavior and accuracy.