Quality Assessment for Stereoscopic Image Based on DCT Frequency Information

摘要

In this paper, we propose an efficient perceptual quality assessment for stereoscopic image based on Discrete Cosine Transform (DCT) coding. In this algorithm, visual sensitivity for human beings to perceive stereoscopic image in frequency domain has been considered. Excepting for this important property related to Human Visual System (HVS), another property is also taken into account: contrast sensitivity function (CSF). In our algorithm, for the left reference block, its most matching block in the left view is found by using the block matching method, and the right reference block is searched based on disparity information as a vital feature in 3D image/video. In addition, its matching block for the right reference one is also finished. The processing mode is completed for the original and distorted stereoscopic images. Then, these blocks are processed by DCT and calculated by modified PSNR in view of CSF and Region of Interest (ROI) mechanism (visual sensitivity). The calculated results have demonstrated its validity compared with the subjective test results. And the proposed perceptual quality assessment metric outperforms other currently popular metrics.