The majority of this post is a copy of an Email I sent to a family friend when I was asked to "digitally restore" some faded and damaged photographs depicting family history which they had scanned for preservation.

The images where quite damaged before the scan, but the Email I received containing 3 Jpegs of these images had far more corruption from the image data compression than anything else, and where stored at a resolution which could be comfortably viewed on screen without reduction.

Admittedly, any higher resolution would only allow you to zoom in on the grainy scratched surface of the image. But it would have lessened the compression errors, and given me a chance to see clean areas between the scratches and grain.

I hope this information will be useful to anyone trying to make digital copies of ageing images (photos, snaps, art work etc) for their preservation.

The illustration above is of two identical images at a resolution of 64 pixels by 64 pixels, depicting a pattern deliberately designed to show up the errors in Jpeg compression.

The left half was saved as a 0.4 kilobyte truecolour PNG image. 32bit, just so there is no argument that the PNG is storing a palletised 6 colour image, where Jpeg will only store grayscale or truecolour. The right half was stored as a 2 kilobyte Jpeg. 4 times the storage space. Each where then blown up to 256 x 256 pixels and aligned alongside each other so you don't have to strain you eyes to see them and the damage caused by Jpeg compression.

Cleaning images is about looking for patterns that are clear from the original, and removing what is out of place. This is made almost impossible by the mess that Jpeg compression makes of regular patterns in images.

In fact, if the image is clean and natural, (where the above example is clearly not) Jpeg does a pretty good job of only removing the patterns our eyes don't notice... the trouble is, when this image is damaged (either by age, or equipment inaccuracy) those areas our brain ignores as regular patterning, are exactly what you need to recreate an estimation of the damaged area.

The Jpeg compression technique (Macro-block Wavelet compression and Discreet Cosign Transformations) is used in DV camcorders (Motion Jpeg), DVD videos (Mpeg) and DVB (h264 / mpeg4)but at least in those forms you have multiple frames with which to smooth out the errors caused by this lossy (messy) compression. Analogue Film, Video and Broadcast also introduce artifacts from analogue compression, but these are of an entirely different nature.

PNG, TGA and TIFF all compress (some optionally) reasonably well, most of the time, without needing to modify the image in order to make it compress better, and frequently compress an image that has previously been compressed as a Jpeg worse (bigger file) than the raw image before Jpeg compression.

If you consider the images above, the one on the left contains many identical repeating patterns of pixel formations, which can be grouped together and replaced with a single <insert here> marker to reduce storage space. The image on the right has destroyed the identical nature of those blocks of pixels, thus destroying it's loss less compressibility. I also mentioned that there where no more than 6 different colours in any of the pixels in the original image, but Jpeg compression has clearly blended, merged and deformed those colours producing many many more shades and hues.

There are very few really good reasons to use Jpeg. Here are a few :-

1. My mobile phone only stores images as Jpeg. (common, sad to say)
2. My web host only accept Jpeg images. (not common, but all sorts of weirdness happens with free hosts)
3. My web browser (or my target audiences) is really really old and can only handle Jpegs, Gifs and the antiquated AOL ART files. (And then, only for full colour images)
4. I really can't afford all the memory cards I would need to store the 500 shots I want to take of this once in a lifetime event in raw format. (And you are willing to sacrifice quality for quantity)
5. My Digital camera is so obscure, I can't find a program that can read it's raw files. (O'Rly? £20 in Tesco will get you a 1.5 MPixel camera which is pretty standard)

Additionally there are a number of times when you should never use any form of image compression which sacrifices image quality for storage space:-

1. For medical photography. (Is that a tumer or a Jpeg artifact?)
2. For scientific photography. (Is there a star in that consolation? IDK could be just compression noise.)
3. For historical archiving. (Have you restored and preserved ageing film? Or have you destroyed the detail it once had with compression noise? Hang on... did Hitler shave for this picture, or has Jpeg just eaten his moustache?)

We are making great strides in restoring film and video footage. With motion mapping you can average out noise and enhance detail, you can even restore lost focus, (sometimes) or interpolate the motion, or resolution between frames.

Consumer demand for High Definition (HD-DVD or Blue-Ray) reproductions of "Frank Cappra's - It's a Wonderful Life" and NASAs increasing frustration with cut-backs forcing astronomical observations back to earth bound measurements are driving us to produce ever more accurate photography by increasing our data set to weed out anomalies.

An very nice technique for non-videographers / cinematographers is called HDR (High Dynamic Range). The process involves taking two pictures in succession, either at different exposures or with and without artificially lighting the subject / scene. (With and without flash, for example) Where you would previously have had to accept an exposure in the middle, which lost some very bright detail and some very dark detail, you can now combine the two images to produce a very natural looking composite which contains great detail both in the very bright and very dark areas.

But sometimes, a single image is all you have. You can't fire an x-ray machine at a man 100 times and average out the differences just to find out what's wrong with him, because the radiation would kill him anyway. You can't go back in time and ask Jesus if he'd mind wiping his brow on a few more shrouds just so we can get a better look at his features either.

If you only have one image to work with, get as much detail out of it as you can. If it's fuzzy or grainy, that doesn't matter. The fuzz and grain is probably in some way related to detail you might just still be able to make some sense out of. If not now, than just wait a while. New ideas are popping up all the time.

Of course C.S.I.s getting a perfect mug shot of a felon off of a hub cap in the distance on a re-used VHS recording from a security camera is complete science fiction. There are too many analogue compression artifacts in that too, and sadly the poor alignment of scan lines in VHS media makes many motion compensation tricks pretty worthless... for now.
Interestingly, if they used Betamax tape, or film the result would be much more plausible, even than a modern DV tape.  O_O

When trying to digitally preserve your aging snaps and photographs, please please please:-

1. Scan big. (Laser printers start at 1200dpi at the bottom end of the range, don't go below that if you can help it)
2. Scan more than once. (You can combine multiple scans to remove artifacts introduced by the scanner mechanism)
3. Scan at more than one brightness and / or contrast setting. (You can increase the range of luminescence in the image that way, so that increasing contrast after the fact does not produce banding)
4. Never, ever, ever compress using lossy methods (Jpeg, Mpeg, Motion Jpeg, JP2000, Gif etc) until you are sure you have finished working on an image, and want to distribute it to people who may not have anything other than a digital photo frame or mobile phone on which to view it.
5. Try to keep a backup of the uncompressed image somewhere. The Jpeg reign of terror has to end sometime soon, and it will be nice to have a clean copy to convert to whatever we standardise on after that. ;)