JPG compression artifacts

[Craig Lamson]

I posted this in reply in another thread but I thought it might have a larger value on it own.

ll be happy to do just that. However I'll offer you the chance to correct this very silly statement of yours first.

Don't wait on my account, Craig ... get to it.

Here you go Bill.

This is the master image. I shot this a few years ago with the original 11mp Canon 1DsmkI. It is a 300dpi file, 12.493" x 8.39" This is a direct png save from my original tif file. The tif and the png match perfectly down to the rgb levels at a single pixel.

http://www.pbase.com/infocusinc/image/90285105/original

Next I created this comp by first opening the original tif file and making 3 duplicate copies. I save each copy directly. One was saved as a png. One was saved as a level 10 jpg. One was saved as a level 5 jpg. and finally the last was saved as a level 2 jpg. These were SINGLE jpg saves. Continued saving of a jpg file causes additional artifacts with each save.

With all of the files saved I created a new document at 300 dpi. I then copied each of the above images and pasted them into this new document and cropped them.

With the crops in place I flattened the image and added the type. The final images was then saved as a png file. There are no addtional artifacts created by saving in png format.

This is the completed composite file.

http://www.pbase.com/infocusinc/image/90285099

When you look at this image at 100%, all the different crops really look pretty close to the naked eye. But if you drill down into the image you can clearly see the destructive effects of jpg compression. The best place to look is the back corner of the mouth. Look at the teeth and then to the redish tones in the roof of the mouth. View the image at 300 to 500%. The easy way to do this is to simply do a right mouse click and then a "save picture as" Be sure to select png as the file format. You can then open this image in any image editting program.

Please note this is a very high resolution example.

[Craig Lamson]

I just noticed that the link to the composite photo leads to the "large" version of this file. Pbase makes different sizes of each posted file and then saves them as a jpg....my bad.

To see the uncompressed original make sure you select "original" from the selection below the photo.

[Jack White]

Nobody doubts the "lossiness" of JPG, as some call it.

However, regardless of the dpi of the original scan, the

internet standard transmission is 72 dpi. So the only way

to increase quality and minimize compression artifacts

sent online is to SEND LARGER IMAGES, since the

www compresses all image transmissions to 72 dpi.

regardless of original dpi scan. However, hi res images

CAN be sent on the www as TIFF files and they will

retain the original dpi even though sent at 72 dpi,

since hi res scans are MUCH LARGER in size.

Jack

[Bill Miller]

Nobody doubts the "lossiness" of JPG, as some call it.

Jack

That's right, but Craig has maintained that Badge Man was created out of nothing. Craig did not show how he took a photo showing nothing there and by the same means as he has said it was done to get the Badge Man image ... Craig did not show how you could make a fish appear that was never there in the first place.

[Craig Lamson]

Nobody doubts the "lossiness" of JPG, as some call it.

However, regardless of the dpi of the original scan, the

internet standard transmission is 72 dpi. So the only way

to increase quality and minimize compression artifacts

sent online is to SEND LARGER IMAGES, since the

www compresses all image transmissions to 72 dpi.

regardless of original dpi scan. However, hi res images

CAN be sent on the www as TIFF files and they will

retain the original dpi even though sent at 72 dpi,

since hi res scans are MUCH LARGER in size.

Jack

Bill Miller doubted it, thus this demonstration. Not all browsers can display tif files, however all will display png format files which is lossless compression.

Edited December 14, 2007 by Craig Lamson

[Craig Lamson]

Nobody doubts the "lossiness" of JPG, as some call it.

Jack

That's right, but Craig has maintained that Badge Man was created out of nothing. Craig did not show how he took a photo showing nothing there and by the same means as he has said it was done to get the Badge Man image ... Craig did not show how you could make a fish appear that was never there in the first place.

Your ignorance is showing now Bill...and your memory loss....

Craig ... scan an image of Badge Man out of a book and post it in Jpeg and then in the format of your preference so we can see this vast difference for study ... I say that the image will be the same - prove me wrong!!!

Looks like YOU doubted it, and were shown once again to be wrong.

And yes Badgeman was created by film grain added in the repeted copy and print process and by overexposure and excessive contrast. That is a fact. You doubt this as well? If so how do you explain the grain present in the drum scan when Polaroid prints are created by the diffusion process and thus have no film grain?

Finally do you have even the slightest clue what lp/mm is and how it effects detail in a print? Time to buy a clue Bill.

Edited December 14, 2007 by Craig Lamson

[Craig Lamson]

Nobody doubts the "lossiness" of JPG, as some call it.

However, regardless of the dpi of the original scan, the

internet standard transmission is 72 dpi. So the only way

to increase quality and minimize compression artifacts

sent online is to SEND LARGER IMAGES, since the

www compresses all image transmissions to 72 dpi.

regardless of original dpi scan. However, hi res images

CAN be sent on the www as TIFF files and they will

retain the original dpi even though sent at 72 dpi,

since hi res scans are MUCH LARGER in size.

Jack

Jack, the web does not compress an image down to 72 dpi, it simply resizes the image without resampling. For example my full image of the fish, if copied from my pbase page will and the opened in a photo editor will display at 72dpi....but a 52"x32". By doing a resize to 300dpi without resampling, brings the image back to its original state with no detail loss which is 12.5"x8.4"

Now I'm not a web expert and there are some sites that MIGHT resize and resample to maintain a certain spec of images displayed on the site. How ever you can post images elsewhere and then link to maintain your original image specs if that is your goal.

[John Dolva]

Craig, this is the type of topic that is much appreciated. The educational content is most helpful.

One matter: some, perhaps all, programs, while resizing, up or down, images (preposting et.c.) will offer some kind of algorithm like bilinear interpolation, pixel resize, Wiener filters, bicubic, bicubic spline, nearest neighbour, cardinal sine, fractal interpolation et.c. et.c. and each introduces different artefacts (often 'invisible' in macro view, but certainly distinguishable when zooming in.) changing the original. A tiny portion of a small polaroid over generational processing invariably introduces artefacts of some kind and a cautioning re drawing any conclusion from the possible alterations by such processing is apt. (IMO)

Edited December 16, 2007 by John Dolva

[Chris Davidson]

John/Craig

Are you familiar with this program and its CDC 4X method?

If so, what is your opinion of it.

This is what I use for enlargements.

Haven't come across anything better that I know of.

thanks

chris

[Craig Lamson]

John/Craig

Are you familiar with this program and its CDC 4X method?

If so, what is your opinion of it.

This is what I use for enlargements.

Haven't come across anything better that I know of.

thanks

chris

No Chris, I've not used this program. I don't upsample often, my current camera gives me 11x17 at 300dpi naive and that is generally all I need for most applications. When I do upsample I use either Genuine Fractals of bicubic in Photoshop in 10% increments with a small USM every third upsample. I have created a photoshop action for this. Both methods have their place.

Maybe when time permits I'll Demo this software.

Thanks

[John Dolva]

Chris, I'm not familiar with that one.

For each type of program there are probably proprietary algorithms that are named differently or the same but yet may in some way be different.

However, I'm using a photoshop clone, and bicubic or bilinear seems best for that proggie.

______

ImageAnalyzer* is well worth a look, it's is free yet very useful. So useful that it doesn't have a comprehensive HELP file, rather one that is put together by users over time, I don't think the writer of the program has fully explored all possible uses. However on his website he discusses resizing and his thinking behind methods.

A couple of features like the fractal interpolation** plugin is good, as well as the 3D plugin which, when one sets the depth map scaling to zero and uses the Gouraud rendering for the texture, one can zoom in and out smoothly to an extreeme degree and take a snapshot which is stored in the main window, ie take numerous snapshots and then choose the one that suits. This one I like, however as my knowledge increases through this forum and through experimenting (eg create a small grid of black and white pixels and try each different algorithm and compare the results) I have become much more sceptical/discerning about borderline issues.

Craig, your technique sounds worth trying. Generally I've assumed that increasing or decreasing by 2X (or multiple thereof) or halving suits most algorithms best simply by the maths involved.

Ultimately (IMO) the best is to discard pre judice and to appraise each equanimously.

A heavy smoking friend of mine got greater results by cleaning the yellowish deposit off his screen. Cleaning the screen regularly is also a good thing. A large fine dot pitch flat CRT is probably still best (unless you can afford the remarkable TV studio graphics screens), but the LCD's are coming along.

* http://meesoft.logicnet.dk/Analyzer/ "There is currently no complete documentation for Image Analyzer, and there probably never will be. Some of the features are described with examples in this page. It is a work in progress, and answers to questions I receive might be added.

For more help, go to our support forum. Everybody is welcome to both ask and answer questions."

** http://meesoft.logicnet.dk/Analyzer/plugin...ctInterpolation - "Fractal interpolation is an alternative to the interpolation methods available in the standard Resize dialog. Fractal interpolation will create artificial details and thus enlarge the picture without smoothing it.

An example comparison is shown in the help. Note this will only give good result on images with little noise, because any noise will be enhanced in the process. It may be advantageous to use the Adaptive noise reduction filter before or after fractal interpolation."

Experiment, compare, read, judge impartially, with good vision.

_________

EDIT : - addendum: with the 3D plugin ( for ImageAnalyzer ) if one (after making a gratscale copy (if the original is not) to use as depth map, with the non-grayscale copy as texture) then setting the scaling at 0.01 or so, one MAY percieve possible areas of retouching ( ~ ), particularly by right clicking to launch the light source and moving it about.

~ correction : one MAY perccieve anomalies that beg an explanation.

Edited December 18, 2007 by John Dolva