What's wrong with this picture?

[Havemercy]

I found this (non-AW) pic buried in the recesses of my hard drive tonight. Is it my imagination, or does there appear to be a gridline pattern over the whole thing? If I zoom in on a section, I see a neat series of white squares. What's going on here? Is this what professional photographers call noise? I tried converting to grayscale, and it's just as noticeable. It's a great photo otherwise.

HM

[Vid Dude]

There are a few possible explanations.

One is it was scanned from a magazine, which sometimes does this moire pattern that can be reduced when the image is resized to smaller than the scan.

Another possibility is that the jpeg has been compressed several times over, generating compression artifacts.

But I think that it is deliberate. A filter has been run over the image to give it some kind of arty style.

[alleyes1]

If that is arty style, I could probably do better with a month of practice.

Alleyes

[Havemercy]

But I think that it is deliberate. A filter has been run over the image to give it some kind of arty style.

Thanks for the info, Vid Dude. Should have asked this earlier, but is there a way to remove or counteract the filter? Very distracting and not the least bit arty, imo. I am not talented enough or rich enough to use Photoshop, but I have something called Graphic Workshop Professional for scaling, cropping, rotating, etc.

HM

[Vid Dude]

No, you can't remove a filter without messing other things up. The clarity would be lost in your attempt to remove something so subtle.

[artlover]

Too bad she's not an Abby girl--filtered or not, she is a beauty. Why try to enhance what needs no enhancement?

[Abby]

Hey, might be texture on the paper that was picked up when it was scanned, too. Weird to print a high detail photo like that on textured paper, tho.

a

[Diablo]

The "noise grid" appears to be aligned on 8x8 pixel boundaries; this looks to me like JPEG quantisation noise instead.

You'll get this if the JPEG encoder is forced to discard/round off too many DCT frequency component coefficients (due to an excessively low quality setting); the error margin at the edge of each 8x8 pixel block becomes significant enough that they fail to seamlessly tesselate with neighbouring pixel blocks.

If what I've written above is all techno mumbo-jumbo to you, the Wikipedia article on the JPEG image format may be of some useful background...

[Vid Dude]

One of these has jpeg artifacting, one has a filter. Which do you think is the closer match?

[Luxman]

The second one. Compressione artifacts give a different result as seen in the first picture.

Amazing: I was just doing the same with a picture when your post came up.

Lxm

[Havemercy]

...Which do you think is the closer match?

You mean some folks leave compression artifacts purposely? I've always heard that too much compression/expansion/copying ruins a jpeg, but I never knew that's what the "blocks" indicated.

they fail to seamlessly tessalate with neighbouring pixel blocks.

Unlike most mathematics, I actually enjoyed geometry, so this makes perfect sense. Thanks all for making just a little bit smarter before bedtime. Look ma, porn is educational.

HM

[Diablo]

Vid Dude: Out of interest, what sort of filter was used on the second image?

A problem I can see with this comparison is the amount of information used to represent each image. The first image is only ~11kB in size, the second is ~31kB. Of course, the first image is going to look worse because you're only using one-third bandwidth. I feel a more meaningful comparison would be to encode both JPEG files so the resulting file size is the same, but apply the filter on one of them prior to encoding (and see if the filter causes the JPEG encoder to quantise the image in a manner more favourable to the human eye).

Havemercy: In theory, if the image has been processed by a filter, you can make a guess at what the filter did, and then produce an inverse transformation filter based on that guess. This would be able to reconstruct the original image to some degree, but the signal-to-noise ratio (SNR) would be worse than the original, due to quantisation of certain signal components in the image that were attenuated by the original filter.

If this is all gobbledegook to you, I can offer a visual demonstration if needed

[Luxman]

Of course, the first image is going to look worse because you're only using one-third bandwidth. I feel a more meaningful comparison would be to encode both JPEG files so the resulting file size is the same, but apply the filter on one of them prior to encoding (and see if the filter causes the JPEG encoder to quantise the image in a manner more favourable to the human eye).

Absolutely correct, but if you want to show compression artifacts (what Vid Dude wanted to do here), you need to compress.

Lxm

[Havemercy]

if the image has been processed by a filter, you can make a guess at what the filter did, and then produce an inverse transformation filter based on that guess.

Makes sense, but as a VidDude said, there's no good way to restore the original sans filter, without ruining the image. And isn't the point of filters and all sorts of special effects to use them in such way that the casual viewer, i.e. yours truly, won't be the wiser? Of course, some effects are quite noticeable if they emphasise a feature, create a mood,et cetrea, but in my pic it just looks like a flat out mistake. I can't imagine why someone would create that effect purposely.

Out of curiosity, under what circumstances would jpeg artifacting be desireable? Again, just looks like a mistake or unfortunate side effect to me. Unless VidDude uses it for demonstration purposes.

HM

[Vid Dude]

Vid Dude: Out of interest, what sort of filter was used on the second image?

It's a mosaic filter in Photoshop. But it's not an exact match with the original picture, so I think that one must've been done with a similar but different filter by a different image program.

As ably pointed out by my good friend Luxman, who is quite knowledgable about digital filters, what I was trying to do (in an extreme fashion) was demonstrate what jpeg artifacting really looks like, which is it muddies and pixelates the image, and removes detail, which is quite different to what the original picture has. Also, the grid pattern is far too evenly applied to be anything other than a filter.

[Diablo]

Oops... scratch my criticism about Vid Dude's comparison... I missed the point, sorry

After some further experimentation, I've found it's possible to supply a JPEG encoder with a custom quantisation matrix that produces the same effect as the mosaic filter in Vid Dude's second example pic. Looking at the attached picture and text file, the matrix I've created simply induces significant error in the two lowest-order AC components of each DCT block, creating that "wavy chunky" effect. The JPEG encoder I used is the Independent JPEG group reference implementation, which lets you tweak many encoding parameters (including supplying custom quantisation matrices).

However whether a sane JPEG encoder would knowingly do that is another story

The noise in havemercy's original pic, while looking uniform at arm's length, loses most of its consistency when you zoom in; it's especially noticeable if you stretch the contrast of the image to maximum. It'd be very difficult to construct a recovery filter that won't decimate the medium-to-high frequency components of the image, or introduce new artifacts.

It still strikes me as odd though that the noise pattern lines up well with an 8x8 grid; perhaps it could be artifacts arising from the scanner, especially if it's attempting to use oversampling by shifting the head in sub-pixel steps. Or something related.

[stoneyyy]

This thread is getting very interesting but I would hate to steal all the thunder and tell you what really happened to that picture havemercy posted..

It was part of and the start of a "dressing room" type of shoot and the cameraman was hiding outside of the window... the patten you see is the flyscreen..

[Dekoda]

I just found this thread, so apologizes for replying so late.

That picture that Havemercy posted suffers from moire. I know....I've scanned thousands of pictures and many of them had that effect. There are moire filters that will fix that but they will cause the picture to lose some detail. I used a moire filter plugin for Photoshop to eliminate the moire, but I had to run the filter twice...once for the horizontal lines and once for the vertical lines. The filter removed the crosshatch pattern, but some detail was lost. Here is the end result.

[Luxman]

Hey Dekoda, a reply after more than 7 years...
I'm impressed!

Lxm

[Dekoda]

Hi there Lux! I didn't see this thread until today. When I saw the original picture that was posted by Havemercy, I knew exactly what was wrong with it, so I decided....after seven years....to reply. I doubt that Havemercy is still a member here, but I couldn't resist replying to his original post as the problem with the picture was something that I am all too familiar with.

[remigius]

See more of ths young lady HERE