Feature Based No-Reference Perceptual Depth Assessment Model for Mobile 3D Video Applications

Abstract

Depth perception is one of the most important characteristics which separate 3D videos from traditional 2D videos. In this work, a feature based no-reference perceptual depth assessment model has been proposed for symmetric and asymmetric coded stereoscopic videos. This model extracts disparity and temporal features to evaluate the perceived depth of mobile 3D videos. The disparity feature is estimated by using block based structural similarity index between the corresponding blocks of left and right view and for temporal feature the jerkiness is estimated between the consecutive frames for both left and right view. The estimated features are then combined to give a single predicted score. The performance of the model is verified by subjective experiment data. The result indicates that the prediction performance of the proposed model is satisfactory.