From H.264 to Hevc: Coding Gain Predicted by Objective Video Quality Assessment Models

Significant progress has been made recently towards the next generation video coding standard by the Joint Collaborative Team on Video Coding (JCT-VC) of ISO/IEC Moving Picture Experts Group (MPEG) and ITU-T Video Coding Experts Group (VCEG). The test model, known as High Efficiency Video Coding (HEVC), aims to achieve 50% coding gain over the current MPEG4/H.264 AVC standard. Recently reported preliminary subjective tests, conducted by JCT-VC members, show that the HEVC draft codec HM5.0 achieves an average of more than 50% rate savings over the H.264 JM18.3 codec without sacrificing subjective quality. In this work, we study the performance of wellknown objective video quality assessment models in predicting subjective quality and the coding gain estimated using the subjective quality scores. We find that state-of-the-art video quality models, including the Structural Similarity index (SSIM), the Multi-Scale SSIM index (MS-SSIM), the Video Quality Metric (VQM), and the MOtion-based Video Integrity Evaluation index (MOVIE), all provide significantly better predictions of subjective video quality than peak signal-to-noise ratio (PSNR), and MS-SSIM achieves the best overall performance. Surprisingly, compared with the coding gain estimated based on the subjective evaluation scores, all the objective quality models involved in the current study systematically underestimate the coding gain of HEVC-HM5.0 upon H.264-JM18.3. We discuss several possible explanations of this somewhat unexpected phenomenon, which provides new insights for the future development of both subjective and objective video quality assessment methods.