Compression's back in a Blaze of glory
The history of digital video compression is littered with companies that promise investors miracles but instead deliver pain.
Former computer games maker Blaze claims it has found the holy grail of video compression - broadcast-quality video down a low-bandwidth link, such as to a GPRS mobile phone. The promise is file sizes can be reduced by up to 90 per cent with a standard-definition broadcast video-quality metric - a standardised performance test for digital video quality.
To put that into perspective, a standard 42-minute TV show occupies up to about 1GB on a DVD, and compresses to about 350MB using Xvid compression. The newer H.264 video compression format reduces this to about 150MB with little noticeable deterioration in quality. Coupled to Blaze's compression method would typically reduce the file size by up to nearly a half - 80-150MB - and its inventors are hopeful that will improve.
Most video compression compares frames and deletes information that does not change, although quality degrades as more information is deleted. Blaze's PIXe (PIXel grid oscillation) technology claims to take advantage of an optical principal called "short-range apparent motion", eliminating image data while fooling the eye into believing it sees information not there.
The technology's developer, Perth-based graphic artist Peter Bevan, says he discovered the concept while considering how the eye eliminates random imperfections, known as "grain", that naturally occur when blowing up film. While a single frame might have so much grain as to be unusable as a still image, its random positioning renders it invisible when projected in sequence.
Mr Bevan, a Blaze non-executive director, says the PIXe process subsamples a video motion picture down to a lower resolution, compressing nine pixels into one, and then uses an algorithm to move the pixel grid around within a frame. The eye averages three or four consecutive frames into one, giving the image definition. Mr Bevan calls this the "grid oscillation process". "We can run film at a lower resolution and reach fidelity levels which are similar to high-resolution film," he says.
"You don't lose much in the look of it - it stays within the broadcast quality ranges - but it is actually quite lower in resolution. Your eye averages it all out and gives you the definition. But if you stopped the film and looked at it, you would have quite a low-resolution frame. The definition doesn't come out until you see it in motion."
A similar technique, "jiggling", avoids pixellation when digital images are blown up for use in billboards. A special camera with an oscillating lens takes 50 images in quick succession and produces an average, effectively replacing the area of one pixel with a greater amount of data.
"We're doing the same thing, except instead of combining (the images in post-production), we're using the eye to combine it," he says.
The output then runs through a standard MPEG video encoder - although MPEG attempts to adjust some of PIXe's oscillations.
Blaze CEO Peter Jon Hartshorne says the decrease in effectiveness of the secondary encoder such as MPEG2 or H.264 reduces the compression efficiency.
Mr Hartshorne pitches PIXe at mobile-phone companies because he says it shows video at 15 frames a second over GPRS-based networks.
"The value proposition of the product will never be in the high-end DVD and motion picture quality," Mr Hartshorne says. "I see this product in lower-bandwidth-constrained applications where you just can't get video to run to an acceptable level. And the most evident market is the low-bandwidth mobile phone."
He says the technology got a positive response from companies with stakes in digital video, such as Microsoft, Qualcomm, NEC and Motorola - but he would not give referees. PIXe was evaluated by Perth consulting group Atrico but its report is unavailable.
Mr Bevan says that not only is the video file size reduced by 90 per cent, so is the computational requirement - meaning cheaper, cooler-running chips and longer battery life in wireless mobile devices such as phones and PDAs.
Blaze engaged US consulting group Gramercy Venture Advisors to organise introductions to clients. Gramercy CEO Michael Gale says scepticism is understandable.
"The problem with compression is there are more kooks per square inch than in anything else," he says. "So far nobody has found any reason why this shouldn't result in a big improvement. Most people are worried that it won't work properly with MPEG, or that it will work, but you will only get a 50 per cent improvement in the end."
Mr Gale says Gramercy is not claiming the technology works but says the early indications are good, while issues remain as to how well it will work with other encoders and how that is built into a service. "We did our own due diligence and we think that it is a very interesting theory," he says. "The way I describe it is there is a lot of circumstantial evidence but no murder weapon."
Mr Hartshorne hopes to have a demonstration version of PIXe ready by January, and is raising a development team in New Delhi to tune PIXe to run on handset operating systems. Blaze will focus on producing a version that works under H.264 MPEG, but has demonstrated its software working with MPEG2 and Motion JPEG encoders, he says.
The Blaze CEO says PIXe needs a proprietary 38Kbps decoder to be downloaded to the handset but the goal is to compress and decode the signal before it is MPEG-decoded.
Blaze has invested $1 million in PIXe, and will invest a further $1 million to get the technology to a demonstrable stage, he says. Two years and a further $5 million-$10 million will be required for a fully commercial product.