extensible cluster frame e remix by eyekhan
Can anyone explain why destroying through remixing a digital image produces all these artifacts? For instance, why does the text at the bottom just spit up all those pixels? Or why did a boundary just start to appear that was never in the original image?
I'll take a stab at it. This image, which is an intermediate image in the process, is a jpeg.
(Techno-babble: While gifs acheive compression via run-length coding and via simplifying the color space, jpegs use a different method. The images are compressed by dividing it into blocks, transforming spatial information into spatial frequency information, and quantizing it.)
JPEG compression has the effect of spatially smearing out the energy in the original image. The smearing around the letters is hard to percieve, because the "white space" around the letters is "almost white". It looks like the remix applied false colors to pixel values. Thus a subtle difference in original color may show up as a radically different color.
So Mark, the "almost white" is the anti-aliasing around the text?
Not quite anti-aliasing, but similar in that it represents a smearing function. Quantization noise is the precise term. It manifests as more of a splotchy smearing than a smooth smearing.
I forced the colors to change but what was weird is that the white around the letters is actually different shades instead of a uniform field.
So is this 'noise' something that jpegs will eventually filter out or is it a permanent flaw?
It is still hard to understand what is happening to the non-tech person like me. Is there anything out there that explains this stuff. I don't even know what anti-aliasing is and how it would relate to this.
Okay, here's an attempt at a simple explanation of anti-aliasing and jpeg quantization. I started with the image below: the letter in in Times New Roman. The letter is anti-aliased, which gives it nice smooth edges.
The detail below shows how the smooth edges are acheived. Rather than an abrupt transition from black to white, there are various shades of grey pixels to create the illusion of smooth curves. This techniqe is very common in digital displays and printing.
The image below is detail of the same letter having passed through jpeg at a high compression ratio. Jpeg compression introduces what is known as "quantization noise". In jpeg compression, the pixels are divided into blocks, and the blocks of pixels are decomposed into a set of wave representations, which are then approximated (quantized). When a jpeg is displayed, the quantized waves are converted back into pixels. Because of the decomposition and approximation process, blockiness, wave-like spotches, and splatters are introduced into the image.
This looks like this is another occassion where i can plug my article
"How to correctly print low resolution screen graphics"
Mark, nice explanation!
Nice article, thanks.
This all started because I was making some little jokes (extremely little).
The image that said ".bmp" was a .GIF that had been made of a capture of a .jpeg. Ha ha ha *slaps thigh.*
The only reason there was a horrible jpeg anywhere in the chain is that Windows XP uses jpegs to make thumbnails of every image on the hard drive (potentially) and uses jpegs to keep the sizes small. What I captured was an image from "My Computer" that said bmp but was in a fact a jpeg. Sort of a slim Magrittean joke. Mainly I was just interested in what happened to the image of the molecule when it was reduced in size and then blown back up, and just sort of rolled with the artifacts, which e emphasized even more with his GIF, to the point where they overwhelmed the typography and became pleasant colored noise.
Mark's and drx's previous writings on this topic had me up to speed so I could indulge in this humor. Also I had earlier posted Paul Slocum's "bad jpeg" as a cross stitch pattern (his idea--I ran it through KnitPro). Now e has been indoctrinated into the mysteries of bicubic mush aka and/or quantization noise.
The images are cool whether or not one knows the underlying math. I especially like image in the post above this one. I've been dealing with image processing for so many years, it's hard for me not to think about what's going on. Can painters look at others' paintings without visualizing brushes and palette knives?
By the way, bicubic mush and quantization noise both have the potential to introduce splotchy smearing, but are different. Bicubic mush is about smoothing to "improve" an image. Quantization noise is about throwing information away to make files smaller.
I changed aka to and/or. I didn't know bicubic mush had such a precise technical meaning--I thought it was a cute way of referring to the characteristic blotchiness that results from resizing or overprocessing an image. Since "mush" is more pejorative than "smoothing."
Tom sorry for making your references to me a pain. I’ll officially use ‘e’ as my name since I have been using ‘j’ jenghizkhan, John, earcon.
I really like this area of working that I feel I just discovered for myself even though it has been going on for a while on this blog.
Digital media is supposed to be so precise, lining everything up into on/off information with, in theory, unlimited possibilities of resolution. The general impression from the everyday user (like myself) seems to be that digital information does not have these inherent limitations that mark and drx have so well made clear in the above articles/explanations. (They really hit home some ideas for me for the very first time.) (The fact that these limitations seem to be driven by our culture’s desires for more instant gratification is interesting in itself because, in theory, digital information need not have any limitations with unlimited file sizes, etc, right?)
Honestly the joke went completely over my head. Now I realize that all these strange artifacts came out of it.
I also see that what I am trying with the images is the same as I have been doing with sound. (Unfortunately, they only appear as bits and pieces in my attempt to make ‘pop’ songs. More serious explorations coming soon.)
I am reminded of those earlier discussions about whether soft synths can be screwed up…
|
extensible cluster frame e remix by eyekhan
- tom moody 3-03-2007 11:59 pm
Can anyone explain why destroying through remixing a digital image produces all these artifacts? For instance, why does the text at the bottom just spit up all those pixels? Or why did a boundary just start to appear that was never in the original image?
- j(earcon) (guest) 3-04-2007 12:18 am
I'll take a stab at it. This image, which is an intermediate image in the process, is a jpeg.
(Techno-babble: While gifs acheive compression via run-length coding and via simplifying the color space, jpegs use a different method. The images are compressed by dividing it into blocks, transforming spatial information into spatial frequency information, and quantizing it.)
JPEG compression has the effect of spatially smearing out the energy in the original image. The smearing around the letters is hard to percieve, because the "white space" around the letters is "almost white". It looks like the remix applied false colors to pixel values. Thus a subtle difference in original color may show up as a radically different color.
- mark 3-04-2007 1:33 am
So Mark, the "almost white" is the anti-aliasing around the text?
- L.M. 3-04-2007 2:31 am
Not quite anti-aliasing, but similar in that it represents a smearing function. Quantization noise is the precise term. It manifests as more of a splotchy smearing than a smooth smearing.
- mark 3-04-2007 5:03 am
I forced the colors to change but what was weird is that the white around the letters is actually different shades instead of a uniform field.
So is this 'noise' something that jpegs will eventually filter out or is it a permanent flaw?
It is still hard to understand what is happening to the non-tech person like me. Is there anything out there that explains this stuff. I don't even know what anti-aliasing is and how it would relate to this.
- e (guest) 3-04-2007 7:19 am
Okay, here's an attempt at a simple explanation of anti-aliasing and jpeg quantization. I started with the image below: the letter in in Times New Roman. The letter is anti-aliased, which gives it nice smooth edges.
The detail below shows how the smooth edges are acheived. Rather than an abrupt transition from black to white, there are various shades of grey pixels to create the illusion of smooth curves. This techniqe is very common in digital displays and printing.
The image below is detail of the same letter having passed through jpeg at a high compression ratio. Jpeg compression introduces what is known as "quantization noise". In jpeg compression, the pixels are divided into blocks, and the blocks of pixels are decomposed into a set of wave representations, which are then approximated (quantized). When a jpeg is displayed, the quantized waves are converted back into pixels. Because of the decomposition and approximation process, blockiness, wave-like spotches, and splatters are introduced into the image.
- mark 3-04-2007 12:49 pm
This looks like this is another occassion where i can plug my article
"How to correctly print low resolution screen graphics"
Mark, nice explanation!
- drx (guest) 3-04-2007 5:02 pm
Nice article, thanks.
- steve 3-04-2007 5:38 pm
This all started because I was making some little jokes (extremely little).
The image that said ".bmp" was a .GIF that had been made of a capture of a .jpeg. Ha ha ha *slaps thigh.*
The only reason there was a horrible jpeg anywhere in the chain is that Windows XP uses jpegs to make thumbnails of every image on the hard drive (potentially) and uses jpegs to keep the sizes small. What I captured was an image from "My Computer" that said bmp but was in a fact a jpeg. Sort of a slim Magrittean joke. Mainly I was just interested in what happened to the image of the molecule when it was reduced in size and then blown back up, and just sort of rolled with the artifacts, which e emphasized even more with his GIF, to the point where they overwhelmed the typography and became pleasant colored noise.
- tom moody 3-04-2007 6:06 pm
Mark's and drx's previous writings on this topic had me up to speed so I could indulge in this humor. Also I had earlier posted Paul Slocum's "bad jpeg" as a cross stitch pattern (his idea--I ran it through KnitPro). Now e has been indoctrinated into the mysteries of bicubic mush
akaand/or quantization noise.- tom moody 3-04-2007 6:18 pm
The images are cool whether or not one knows the underlying math. I especially like image in the post above this one. I've been dealing with image processing for so many years, it's hard for me not to think about what's going on. Can painters look at others' paintings without visualizing brushes and palette knives?
By the way, bicubic mush and quantization noise both have the potential to introduce splotchy smearing, but are different. Bicubic mush is about smoothing to "improve" an image. Quantization noise is about throwing information away to make files smaller.
- mark 3-04-2007 10:41 pm
I changed aka to and/or. I didn't know bicubic mush had such a precise technical meaning--I thought it was a cute way of referring to the characteristic blotchiness that results from resizing or overprocessing an image. Since "mush" is more pejorative than "smoothing."
- tom moody 3-04-2007 10:49 pm
Tom sorry for making your references to me a pain. I’ll officially use ‘e’ as my name since I have been using ‘j’ jenghizkhan, John, earcon.
- e (guest) 3-04-2007 11:59 pm
I really like this area of working that I feel I just discovered for myself even though it has been going on for a while on this blog.
Digital media is supposed to be so precise, lining everything up into on/off information with, in theory, unlimited possibilities of resolution. The general impression from the everyday user (like myself) seems to be that digital information does not have these inherent limitations that mark and drx have so well made clear in the above articles/explanations. (They really hit home some ideas for me for the very first time.) (The fact that these limitations seem to be driven by our culture’s desires for more instant gratification is interesting in itself because, in theory, digital information need not have any limitations with unlimited file sizes, etc, right?)
Honestly the joke went completely over my head. Now I realize that all these strange artifacts came out of it.
I also see that what I am trying with the images is the same as I have been doing with sound. (Unfortunately, they only appear as bits and pieces in my attempt to make ‘pop’ songs. More serious explorations coming soon.)
I am reminded of those earlier discussions about whether soft synths can be screwed up…
- e (guest) 3-05-2007 12:13 am