Almost Live from Streaming Media East: EuclidIQ Debuts IQ264
[Note: This sponsored interview was recorded at Streaming Media East 2015.]
EuclidIQ has found the secret to creating a better encoder. As Dr. Nigel Lee, the Concord Massachusetts-based company's chief science officer, explained at Streaming Media East 2015, it's all about understanding human perception.
"Traditional encoders maximize quality by minimizing the error between encoded frames and the original frames, but they treat all errors in the frame equally; but we know from experience that this is inaccurate," Lee said. "When human observers watch a video, they tend to notice some parts of the video more than others. Traditional encoders don't take human perception into account; so they might spend too many bits trying to improve quality in less important parts of the video frame."
To solve for that problem, EucidIQ created IQ264 video compression technology, which it showed off for the first time at Streaming Media East 2015.
"Our approach is to apply perceptual quality optimization to the video encoding process. We create what we call Unified Importance Maps that combine both spatial and temporal information to predict what parts of each video frame will be most noticeable to human observers," Lee explained. "We then instruct the encoder to spend more bits and improve quality in the parts of the video that are more noticeable, and to spend fewer bits and reduce encoding size in the areas that are less noticeable."
To learn more about IQ264 and the compression savings it offers, watch the interview below.
Troy: Hi everyone, this is Troy Dreier of StreamingMedia.com, coming to you almost live from Streaming Media East 2015. I'm pleased to talk to some of the news makers of the show here today, and I am joined by Dr. Nigel Lee, Chief Science Officer for EuclidIQ. Welcome to the show.
Nigel: Thank you.
Troy: You were working at the booth, talking to customers, getting some prospects. How's it been going?
Nigel: Pretty well.
Troy: But we're really here to talk about what you guys do, and you guys are compression specialists, right?
Nigel: That's right.
Troy: Can you describe some of the limitations of traditional video encoding?
Nigel: So, Troy, all video encoders seek to encode videos with the smallest size possible, but with the best quality possible. Now, it's not possible to optimize both size and quality at the same time, so different encoders try to trade off one for the other in different ways. Traditional encoders maximize quality by minimizing the error between encoded frames, and the original frames, but they treat all errors in the frame equally; but we know from experience that this is inaccurate. When human observers watch a video, they tend to notice some parts of the video more than others. Traditional encoders don't take human perception into account; so they might spend too many bits trying to improve quality in less important parts of the video frame, or they might spend too few bits, trying to reduce encoding size in more important parts of the video frame.
Troy: But you guys are different. You guys have come up with a solution to that.
Nigel: That's right. Our approach is to apply perceptual quality optimization to the video encoding process. We create what we call "Unified Importance Maps" that combine both spatial and temporal information to predict what parts of each video frame will be most noticeable to human observers. We then instruct the encoder to spend more bits, and improve quality in the parts of the video that are more noticeable, and to spend fewer bits, and reduce encoding size in the areas that are less noticeable. Doing this results in improved compression performance, relative to traditional encoders.
Troy: You've created a way to automate what's the most important thing in the frame. What does the encoder look for, what is the human eye drawn to?
Nigel: Our approach to perceptual encoding optimization is, at the same time, more sophisticated than some techniques, and more flexible than others. Some encoders apply a very simple approach of adjusting video quality, encoding quality based on block variance as a measure of complexity, with the idea that low variance, or low complexity data should be encoded with high quality. Our approach is more sophisticated than that, because we recognize that some low variance data is more important than others, based on, for example, surrounding data, or time history. Other past, other earlier attempts at perceptual quality optimization applied object detection, and encoded foreground objects differently than the background, now, that object-based approach turned out to be both too complex, and too rigid. Our approach, by contrast, is more flexible, because it does not attempt to detect objects in the data, and it thus can be applied to any motion compensation-based video compression standard, including H.264, HEVC, and VP9.
Troy: Interesting, so what formats is this available in? How do people get this?
Nigel: Our IQ264 compression suite implements our perceptual quality optimization technology for standards-compliant H.264 encoding. We've developed an SDK that allows IQ264 to be used by any software-based H.264 encoder; so far, we've integrated our IQ264 SDK into x264, which is one of the very best performing H.264 encoders, and in double-blind subjective testing, we found that IQ264 on top of x264 produces 20% compression gain, relative to x264, even when x264's most sophisticated compression algorithms are turned on.
Troy: That's pretty impressive, but that's got to add latency to the process, right?
Nigel: Well, our current implementation of IQ264 is fast, and runs at nearly the same speed as x264, so we believe that we have a version of it that is practical, in that sense.
Nigel: If anyone's interested in trying out our IQ264 SDK, they can contact us at email@example.com, or visit our website www.euclidiq.com.
Troy: Fantastic, and everybody here is always looking for a smarter compression, and that sounds like something a lot of people are going to want to take a better look at. Well, thank you for joining me, Dr. Lee; and this is Troy Dreier coming to you almost live, from Streaming Media East.
Nigel: Thank you.
This article is Sponsored Content