Xerox Begins Rolling Out Copier Software Fix

CommanderFrank

Cat Can't Scratch It
Joined
May 9, 2000
Messages
75,399
Xerox has begun the task of updating corrupted software on a wide range of their multifunction printers. The patches will be rolled out starting this week and completed the following week.

The first patch is for the Xerox ConnectKey family, WorkCentre 75xx, WorkCentre 57xx and ColorQube 93xx machines. A second patch will be available next week that covers the rest of the affected products.
 
Yes, it's possibly compression related, but also: crap in, crap out.

Nope, today's news report is about Xerox's patch that fixes the problem entirely. As in, doesn't scramble the page it's scanning by moving bits and pieces of the document to other parts of the document.
 
Nope, today's news report is about Xerox's patch that fixes the problem entirely. As in, doesn't scramble the page it's scanning by moving bits and pieces of the document to other parts of the document.

The article doesn't go into detail about what caused it in the first place. Is there a link to that info?
 
I wouldn't be suprised if this guy has some unfortunate accident happen to him in the future.
 
The article doesn't go into detail about what caused it in the first place. Is there a link to that info?

The link I posted is the most information about it you'll ever see. Xerox is in full damage control mode, because they could be up for major lawsuits. Did you notice how a construction diagram got scrambled? Think about that - did any buildings get built wrong because of this bug?
 
The link I posted is the most information about it you'll ever see. Xerox is in full damage control mode, because they could be up for major lawsuits. Did you notice how a construction diagram got scrambled? Think about that - did any buildings get built wrong because of this bug?

They shouldn't be building based on low resolution scans. The contractor will have access to the originals, plus this is all done in CAD already. Scans are for archival purposes when it comes to contruction.
 
They shouldn't be building based on low resolution scans. The contractor will have access to the originals, plus this is all done in CAD already. Scans are for archival purposes when it comes to contruction.

You're missing the point which is the fact that this is happening is a serious and dangerous problem. Not one person, myself included ever thought or had to think that their scans would be altered and altered in this way. This is not good.
 
You're missing the point which is the fact that this is happening is a serious and dangerous problem. Not one person, myself included ever thought or had to think that their scans would be altered and altered in this way. This is not good.

No, I get that point. I'd be pissed. I'm just more interested in the details than the fallout.
 
What would possess someone to write code that replaces pixels in a scanned image with different numbers?
 
What would possess someone to write code that replaces pixels in a scanned image with different numbers?

They didn't, per se.

What I believe happens is it acts similar to how ZIP compression works, in that it builds a "database" of hashed values with their real outputs. Since the patch/block sizes are pretty large sometimes sections will get scanned in and the algorithm will say, "Yep, these are the same." When it takes the open tail on a '6' and says "that looks enough like the loop on a closed 8", the algorithm saves it that way. When the image is reconstituted for output it simply works off of the algorithm-generated result. Since it was told "this is the upper-right corner of an 8" it prints that out (NOTE: it isn't actually "thinking" with numbers/letters, it's just encoding a digital snapshot of that portion of the output, unaware that what it's encoding may be interpreted as letters/numbers/symbols.)
 
They didn't, per se.

What I believe happens is it acts similar to how ZIP compression works, in that it builds a "database" of hashed values with their real outputs. Since the patch/block sizes are pretty large sometimes sections will get scanned in and the algorithm will say, "Yep, these are the same." When it takes the open tail on a '6' and says "that looks enough like the loop on a closed 8", the algorithm saves it that way. When the image is reconstituted for output it simply works off of the algorithm-generated result. Since it was told "this is the upper-right corner of an 8" it prints that out (NOTE: it isn't actually "thinking" with numbers/letters, it's just encoding a digital snapshot of that portion of the output, unaware that what it's encoding may be interpreted as letters/numbers/symbols.)

No, it is nothing like ZIP compression. ZIP Compression is lossless. This, obviously, is not. Only an idiot would use lossy compression on textual data.
 
They shouldn't be building based on low resolution scans. The contractor will have access to the originals, plus this is all done in CAD already. Scans are for archival purposes when it comes to contruction.

have you ever seen the full chain of building and prints? I'm an apprentice electrician, and I only ever get crappy scanned and photocopy images to go off of. While this is true for the contractors its common for the actual workers to receive Xeroxed copies to go off of as there not going pay that extra to make sure every worker has high end copies when there just gonna get dirtied up and destroyed as time goes on.

Would I like nice fresh high end copies? Sure, but I deal with the photocopy of my journeyman's paper with his scribbles and scratches on it already as its better than walking to him every hour asking ok what do I need to wire next? Im sure not every trade is like this, but as an electrician almost every apprentice/group of apprentices needs their own copies as theres way to much for us to do without it.
 
They didn't, per se.

What I believe happens is it acts similar to how ZIP compression works, in that it builds a "database" of hashed values with their real outputs. Since the patch/block sizes are pretty large sometimes sections will get scanned in and the algorithm will say, "Yep, these are the same." When it takes the open tail on a '6' and says "that looks enough like the loop on a closed 8", the algorithm saves it that way. When the image is reconstituted for output it simply works off of the algorithm-generated result. Since it was told "this is the upper-right corner of an 8" it prints that out (NOTE: it isn't actually "thinking" with numbers/letters, it's just encoding a digital snapshot of that portion of the output, unaware that what it's encoding may be interpreted as letters/numbers/symbols.)

According to the article links in the first reply that pretty much is what they are saying the issue is. If you use OCR or TIFF then you were fine. if you used the normal mode then it used JGIB2 and would use an image database and compare the two. Think the 6 and 8 looked close enough within its margin of error and flip them out.

have you ever seen the full chain of building and prints? I'm an apprentice electrician, and I only ever get crappy scanned and photocopy images to go off of. While this is true for the contractors its common for the actual workers to receive Xeroxed copies to go off of as there not going pay that extra to make sure every worker has high end copies when there just gonna get dirtied up and destroyed as time goes on.

Would I like nice fresh high end copies? Sure, but I deal with the photocopy of my journeyman's paper with his scribbles and scratches on it already as its better than walking to him every hour asking ok what do I need to wire next? Im sure not every trade is like this, but as an electrician almost every apprentice/group of apprentices needs their own copies as theres way to much for us to do without it.

Makes sense to me.
 
Back
Top