SSIMPLUS Outperforms Netlix’s VMAF
SSIMWAVE’s perceptual video quality measure beats video delivery giant’s in-house lab for speed, accuracy and completeness.
Waterloo, Ontario, Canada(31 Jul 2017)
It’s a complex combination of configurations balancing "quality" with the realities of bitrates, encoding parameters, servers, delivery networks, and consumer devices. What matters most is the quality of what the human viewer sees.
A new study from the University of Waterloo (Ontario) found SSIMWave’s SSIMPLUS process is 10 times faster and 49% more accurate than Netflix's VMAF.
Current processes test configurations and outputs with mathematical difference quality measures and human experts aptly named "golden eyes." The process of relying on humans doesn’t scale for today’s explosion of video consumption. The question the industry is asking is "Can we trust computers to the task?" Some would argue never, but SSIMWAVE’s SSIMPLUS and Netflix’s VMAF are out to prove otherwise by narrowing the gap between objective (modeling/predicting/measuring) and subjective (human opinion scores).
Researchers and the industry have been working on closing this gap (See Advancements) but to get computers close to how humans perceive video and for them to work in real-world environments, the solution better be comprehensive, highly accurate, real-time fast, and pay attention to how and where individuals experience moving pictures. The results of this test are proof that future is here with SSIMPLUS.
The best part of industry and researchers striving towards the same goal is that publicly available datasets are available so that anyone can put the different perceptive quality measures to the test against real videos and real people. The comparison shown was conducted by the world-renowned video lab at the University of Waterloo.
Mathematical Difference, Machine Learning and Viewer Intelligence have created significant advances in measuring video quality and virtualizing human viewing experiences.
PSNR is considered the baseline algorithm. It’s a simple ratio between power of a signal and the power of corrupting noise
VMAF Video Multimethod Assessment Fusion introduced by Netflix in 2016 makes use of machine learning to predict subjective quality by combining multiple elementary quality metrics.
SSIMPLUS, introduced by SSIMWAVE Inc. in 2014, makes use of Viewer Intelligence and accounts for behaviours of the human visual system, video resolution, frame-rate, dynamic range, display devices, viewing conditions, network impairments, and viewer type.
We encourage you to run independent tests. The datasets, tools and resources containing reference videos (representing sports, news, action, cinema, animation) and grades (human opinion scores) of hundreds of thousands of real people are all publicly available. Click here to download instructions.