MS Proposes JPEG Alternative
automatix writes "Microsoft's new competitor to the omnipresent JPEG format has been shown at WinHEC and is discussed on CNET. The Windows Media Photo format has many promises associated with it. The program manager is claiming 'We can do it in half the size of a JPEG file.'. While 'the philosophy has been that licensing should not be a restriction', it is interesting that the specification requires a click-through agreement to even read it."
Reading all of 31 pages of the document makes me understand that it is just an attempt to hijack tiff an bend it with MS patented pixel codec to become incompatible with existing tiff technology. Salted with Adobe XMP metadata, ICC metadata and EXIF metadata. All of that registered as a Microsoft trademark. Did I missed something?
There you are, staring at me again.
Jpeg2000 supports both lossy and lossless compression, and a variety of wavelet-based compression schemes that work better than normal JPEG.
Unfortunately, no-one supports Jpeg2000.
Athletic Scholarships to universities make as much sense as academic scholarships to sports teams.
Ok fine no one gets it. A banddand is a range of frequencies you will find in an image. As it turns out, we don't respond to error in an image by the image itself, but by the frequency that the error in the image disrupts. We're worse at seeing disruptions in the high and low range of frequencies, and better in the midrange. Somewhat ironically that means we can take advantage of the high and low and compress more inside those frequency ranges. A DWT or DCT wil give you component pieces for various frequencies which you can simply or delete to form the compression (DCT is JPEG, DWT is JPEG2000). Remember the square blocks in JPEG compression? That's from the DCT. The DWT is more circular so you'll never see square blocking with JPEG2000.
l er_5749_40.pdf
If anyone is interested and wants some not-so-light reading, check out http://foulard.ece.cornell.edu/publications/chand
It'd be awesome if someone made a compressor for regular images using this technique.
I only did one Google search, but easily came up with this old article from last October. I haven't really followed the case, but it's one reason why MS may have done this.
First of all, forget color. A color image is just three channels of black and white.
Imagine you have a back and white image which is pure white noise. Consider what a single horizontal line of that image would look like if you drew it as you would a sound wave, with the bright pixels being high, and the dark pixels being low.
As you step from one pixel to the next, you could have a change of up to 255. There's no predictable pattern. The "frequency" of this noise is high, because the potential difference from one pixel to the next is great.
Now imagine that you apply a smoothing filter to this line of noise, and bring the changes from one pixel down. That is what you get if you blur an image. Now the max differences from one pixel to the next is much lower. The frequencies in a blurry image are low.
There's other ways to consider the frequencies of an image as well. In Wavelets, you would scale the image down to 2x2, and this would be one layer of the image. Then you'd scale it down to 4x4, and scale up the 2x2 image with bilinear filtering and subtract it from the 4x4 image. The 4x4 difference image now represents a different set of frequencies than the 2x2 image did. You store the difference because what you're interested in is the frequency of the 4x4 layer. You want to add that frequency on top of the 2x2 layer when you reconstruct the image, and if you have that "frequency" seperated out, you can compress the data better.
Another way of looking for frequencies in an image is to seperate the image into bitplanes. I think TIFF does this, because it comrpessed the image about the same as seperating the image into bitplanes then compressing with zip. Anyway the idea here is to take all the first bits of each pixel and stick them one after another, and then stick the second bits of all the pixels one after the other... You'll end up with 8 images this way, and you'll find that the image with the highest bits is easily recognizeable and has clear sharp edges, but when you get to the image with the lowest bits, all you have is noise. If you discard that noise when reconstructing the image then you will get banding in the image, but you could in theory interpolate the values of the band above to fill in the noise. You'll lose noise in the image though so stuff will look smoother than it did. Wavelet does somethign similar when it discards the differences and smooths the portions of the image that are in between sharp edges.
If you click on the "I accept this agreement and want to download the Windows Media Photo Specification" button, it submits "I accept this agreement and want to download the Windows Media Photo Specification", and should take you to http://www.microsoft.com/whdc/xps/wmphotodwn.mspx? . However, I didn't verify that.
Instead, I chose to look at the HTML, and manually submitted my own prefered value via manually entering the URL: http://www.microsoft.com/whdc/xps/wmphotodwn.mspx? I_Reject_The_Agreement_Terms_and_Suspect_Bill_Gate s_Blows_Goats. I also got taken to the download page. This page contains the notice "By installing, copying, or otherwise using the software, you agree to be bound by the terms of the license agreement.", and a download link to the actual specification document at http://download.microsoft.com/download/1/6/a/16acc 601-1b7a-42ad-8d4e-4f0aa156ec3e/WMPhotoSpec_v09.do c....
Oops.
Now, while I Am Not A Lawyer, I submitted my rejection of their license terms, so I'd argue in court I shouldn't be bound by them; and since this is a specification, and not itself software, I would also argue that the notice on the page I reached is moot. I suppose the case could be made that since Word macros are a turing-complete programming language, the word document is software, so I thought I'd look through using "less" to be on the safe side. Lo and behold, there is another license embedded:
Of course, if someone at a unix command prompt incanted something clever (say, curl -o Bill_Blows_Goats.txt -C 8261 http://download.microsoft.com/download/1/6/a/16acc 601-1b7a-42ad-8d4e-4f0aa156ec3e/WMPhotoSpec_v09.do c — and don't forget to remove the Slashdot inserted spaces) the Microsoft server would only give them the meaty parts (albeit in a form even OpenOffice would probably gag on), and omit the license. I'd be amused to hear the opinion of a Real Lawyer as to how binding the agreement co
//Information does not want to be free; it wants to breed.