3d-dct Based Multi-scale Full-reference Quality Metric for Stereoscopic Video

In this paper, a new 3D video quality metric based on multiscale images and masking effect is proposed. Proposed metric mainly contains three components: 3D-DCT transform, asymmetric multi-scale measurement, and difference of disparity estimation. 3D-DCT transform is processed by grouping similar blocks in both left and right channel. DCT coefficients are analyzed taking into account Human Visual System (HVS) properties, i.e. contrast sensitivity function (CSF) and masking effect. Left channel is processed into several downsampled versions and then measured by PSNR-HVS-M. In addition, depth error between original image and distorted image is also measured using disparity estimation. The proposed metric is validated over 3D video quality database with wider types of compression artifacts. The results show that the metric outperforms current metrics over different 3D video formats and compression methods.