In the same way that location is the only thing that matters when it comes to real estate, there’s only one thing that’s important when it comes to video delivered over a broadband network: Viewer, viewer, viewer. Why? Because it’s the viewer who determines what gets watched — or not. It’s the viewer who determines how engaging a piece of content is — or isn’t. It’s the viewer who decides if the quality of a signal, and the overall viewing experience, is good or bad. And, ultimately, it’s the viewer who decides who gets paid. Because if the broadcast quality is poor, if the experience unsatisfactory, subscribers will flee from a platform, be it Netflix or Rogers or network TV. “In the end, what matters are the final viewers who are watching video on their phone or their tablet or their TV,” says SSIMWAVE’s Chief Science Officer, Prof. Zhou Wang. “It’s really that simple.” As a result, SSIMWAVE’s products — technology that’s designed to help broadcasters measure and achieve video quality — are built with one goal in mind: viewer satisfaction. The viewer is so fundamental to SSIMWAVE’s mission that its products aim to mimic the human visual system, to see what humans see, to know what the human visual system knows. “The reason is very simple,” says Wang. “A viewer will watch a video and make conclusions about the quality. They’ll say: ‘Excellent,’ or they’ll say, ‘good,’ or they’ll say, ‘bad.’ They’ll say, ‘This feed made my app crash, so it’s terrible.’” “That single word from them — excellent, good, bad, terrible — really defines everything.” “The ultimate goal for us is to optimize what people will say.” What people will say. The problem facing broadcasters and viewers alike is that the journey a video makes enroute from its source, to a distributor, to a device, and ultimately to the viewer’s eye, is one that’s full of bumps and bruises. During the journey, the signal is repeatedly compressed and altered. “It goes through many, many different stages,” says Wang. How can engineers along the delivery chain know the impact those compressions and changes are having on the quality of the signal? How can they succeed in delivering a product to the end of the chain that their viewer will say is of a high quality? How, ultimately, can they see what their viewer sees? “When you know the physics behind the human visual system, you can build something much smarter, and much faster.” One way, says Wang, would be to hire many, many people and place them at every transition point in the delivery chain, monitoring the video and its quality along its journey. Of course, logistically, that would be problematic. There are tens of thousands of feeds that would need monitoring 24/7. Not only would doing so require millions of people, it would be prohibitively expensive. The alternative? SSIMPLUS. “You can replace all these people with our software probe and our quality measurement, the SSIMPLUS Viewer Score,” says Wang. “Our product is already proven to be running all these things in real time and deployed on a large scale at a low cost.” “That’s our business. It’s as simple as that.” Executing on that business, inventing the secret sauce that lives inside SSIMPLUS, required making some deliberate decisions on the viewer’s behalf. In particular, it required not doing video quality measurement the way it had been done in the past by SSIMPLUS’s competitors. What are those methods? One typical, and popular, quality measurement tool used in the broadcast industry is PSNR, or Peak Signal to Noise Ratio. There are others, like it, too — bit error rate, packet drop rate, network delay, frequency of a stalling event — “simple mathematical models,” Wang calls them, and while they’re all somewhat applicable to the task of measuring video quality, they don’t do what SSIMPLUS does. “They’re relevant, but there are also things they don’t consider,” says Wang. “There’s nothing there that can distinguish between whether [a video is] excellent quality, for example, or just OK quality.” Likewise, Netflix uses a quality measurement tool called VMAF, or Video Multimethod Assessment Fusion, in conjunction with machine learning. Wang describes VMAF as an improvement over PSNR. “Netflix cares about video quality a lot,” says Wang, “and VMAF is a good method.” The problem with VMAF, says Wang, is speed. “It’s not that fast,” he says. “SSIMPLUS is 10 times faster than VMAF. It might be even faster.” But there’s more. Neither PSNR nor VMAF is able to predict the differences in Viewer Experience when watching the same video on a TV screen versus a cellphone. SSIMPLUS can do so. And finally, neither VMAF nor PSNR achieves SSIMWAVE’s statistical relevance, namely providing a better than 90 per cent correlation accuracy between SSIMPLUS Viewer Score and subjective human opinion across all video content. In other words, SSIMWAVE’s technology has the ability to see and measure the quality of a video the way a human being sees quality — to serve as a proxy for a person’s eyes along the delivery chain. “We’re trying to mimic the human visual system,” Wang says. “We start with very basic things, pixels, and we build models from the human visual system and directly reflect its most important elements.” “When you know the physics behind the human visual system, you can build something much smarter, and much faster.” SSIMPLUS assesses video quality and then presents its findings in real time in the form of a relevant, easy-to-read dashboard with one score simplicity, allowing engineers to see in an instant whether quality is being compromised. If problems with a video transmission arise, the dashboard allows technicians to quickly drill down and discover where, and why, and what remedial actions they should take. “With SSIMPLUS, we have two goals,” says Wang: “Improve the correlation with human perception of image and video quality, and to make it extremely fast. It has to be fast enough that it costs very little for very large-scale deployment in terms of computation.” Because in the end, what matters isn’t a mathematically rendered error rate, or bit rate, or freeze rate. What matters is what a person at the end of the line thinks. What matters is what the viewer sees on their device or television. What matters is that their viewing experience is reliable enough, transparent enough, that they’re free to be swept away by the emotion of a film, or an event, or a game, or a moment. At SSIMWAVE, what matters is viewer, viewer, viewer.