No-Reference DIBR-Synthesized Video Quality Assessment based on Spatio-Temporal Texture Inconsistency Measurement

Abstract

The relevant applications of depth-image-based-rendering (DIBR) exist mainly in the form of video sequences. However, existing studies on the quality assessment of DIBR-synthesized views primarily focused on DIBR-synthesized images. To this end, this paper proposes a DIBR-synthesized video quality evaluation metric based on measuring spatio-temporal texture inconsistency, dubbed STTI. Specifically, STTI first extracts the texture map of each frame in the spatial domain. Then, STTI further employs the histogram of oriented optical flow to extract the dynamic variations of adjacent frames' texture information in the spatio-temporal domain. Finally, STTI calculates the cosine similarity of the histograms of oriented optical flow between the texture maps of adjacent frames to measure spatio-temporal texture inconsistency. Experimental results on the publicly available datasets show that the proposed STTI outperforms the popular image/video quality assessment methods developed for natural scene and DIBR-synthesized views.