Software that minimally degrades JPEG files upon saving?

To illustrate what I am talking about, I'm going to attempt to attach a .gif illustration of a 640x512 image, with 80 MCUs. Each MCU has 64 Y 8x8 DCTs, each representing 8x8 pixel blocks, and 1 each of a U and V 8x8 DCT, each representing 64x64 pixel blocks. (Crude, but I think it would comply with the JPEG standard--but again, it's been awhile).

The problem I am hearing with your reasoning is not that it could not work, but that it is too low level of an approach to the problem. I have seen references to JPEG editing software that achieves this type of result, but it is of limited use as it only works on jpegs and no professional with any integrity would send jpegs to press if they can avoid it.

Next you are assuming that when PS (Photoshop) opens a JPEG that it retains the fact that the data came from a JPEG. Why would you do this when your final output data type is unknown (TIFF, TGA, PSD, EPS, ...) and may not even be a JPEG? So far as I can tell, PS only tracks enough data to create any valid output type. Hence, once an image is loaded the only data stored is that about the image itself and not about a specific compression method. This is changing with the introduction of EXIF data, but even that extra data is about the image and not about how it is stored.

As a result, I think you are looking at this problem from the wrong direction. The only time we wish to retain the extraneous details about JPEG compression is when we are dealing with the JPEG directly. Otherwise we just wish to deal with the image data. Hence every time PS's JPEG compression algorithm comes into play it assumes it is dealing with uncompressed data and starts from scratch.

On top of that, there is the issue of the quantization tables used by various JPEG implementations. To retain the data you wish to retain, every implementation must use the same quantization tables even though this is an implementation defined feature. Plus, not all quantization tables are created equal. Some quantization tables may be better suited to certain image classes (say photos of feathers or photos of gemstones). Hence tracking this info when the original file is using quantization tables good for gemstones may be a very bad idea if your user is working on photos of horses may be an awful idea.

I am sorry if this is not 100% clear, but without body language to see what you understand this is a tough thing to explain. Most of the reasoning behind this that I can ferret out is based on higher level decisions regarding software architecture.

In short, JPEG is a way of storing images. All JPEGs are images. But not all images are JPEGs. PS is a program for manipulating images. Hence it would be very silly to calculate these changes to the jpeg when they alreasy have a 100% accurate map of the changes in their internal representation of the image itself.

hope this helps,

Sean

I wrote an article on Jpeg Compression (http://www.photo.net/learn/jpeg/index.html) that covers most of the issues you mention. As to why commercial software doesn't do this - your guess is as good as mine...

In theory, nice. But in reality, the reason I open and edit images in Photoshop is to asjust the whole colour balance of the image, then resize and save again. So the whole image changes and I doubt many people _just_ edit parts of images, in which case that part would still degrade.

Thanks for the responses.

Sean De Merchant wrote:

>I have seen references to JPEG editing software that achieves this >type of result, but it is of limited use as it only works on jpegs >and no professional with any integrity would send jpegs to press if >they can avoid it.

I agree that it is of limited use to professionals. But it is of great value to the millions of non-professionals who are buying P&S digital cameras and using their old computers with limited storage capacity. It is also of value to people who inadvertently resave a .jpg file from their digital cameras. Do you recall which software does this? I'd like to experiment with a demo version if available.

>Next you are assuming that when PS (Photoshop) opens a JPEG that it retains the fact that the data came from a JPEG.

This is a safe assumption. When I hit <cntrl>-s, it automatically saves an open jpeg as a jpeg (unless I've done something fancy like add layers). It recalculates a whole slew of DCTs unnecessarily, too, considering that for a given unedited MCU, its DCTs are already in the original .jpg file. Using a scorecard would reduce these unnecessary calculations (you probably wouldn't notice the speed up compared to FILE I/O bottlenecks) and may increase the amount of memory used.

>Hence, once an image is loaded the only data stored is that about the image itself and not about a specific compression method.

Yes, I think this is how they handle it, and I suspect other image editing software packages do the same, but it is not the most clever way of handling it. Or maybe it is...

Mr. Richardson wrote:

>As to why commercial software doesn't do this - your guess is as good as mine...

Since the original post, I was thinking/guessing that one valid reason might be that they don't want higher quality MCUs stored in an image with degraded MCUs, which might become annoying to users and/or technical support. Better to degrade the whole image than just the section touched by the tool (PS, etc.). Who knows.

Anyway, I've made another more realistic .gif image of MCU mappings in JPEG. See attached.

For jpeg editing take a look at:

http://www.pegasusimaging.com/jpegwizard.htm

As for other issues. Yes, PS does resave a JPEG as a JPEG, but this info is retained at file level. A TIFF is implicitly resaved as a TIFF.

As for disk space, you can pick up a CDRW for about $20 US on sale and 50 blank CDRS for another $10 US on sale. This gives you an additional 35 GB of storage for about $30 US. After spending $600+ US on PS, this is a drop in the bucket. I know I have many many GB of images on CD and I am contemplating the upgrade to DVD+R or getting a high capacity tape backup. But for now my $20 US 52X CDRW will do the job.

enjoy,

Sean

Most Jpeg coders and decoders are nearly symmetrical (IJG are), and so the loss in repeated saves is minimal. It is not necessary to go down to such technical level - it is sufficient to keep the same quantisation tables (and chroma subsampling). This can easily be done using the IJG program CJpeg.Exe with the "-table" parameter, which then reproduces the original Jpeg almost exactly.

There is also a program called JpegCrop.Exe (from http://www.jpegclub.org/) which can loslessly rotate and crop Jpeg's by rearranging the encoded bits.

It is not clear what problem you are trying to solve, other than the trivial one of opening and closing a Jpeg without making any changes? Editing a Jpeg is not a good idea in principle (although it practice it is not as harmful as it seems), so encouraging this concept is probably counterproductive. A popup warning when overwriting an existing Jpeg might be more constructive.

Here are some sample 8X8 quantisation tables (luminance only). The tables from any given Jpeg can be extracted with the program JpegDump.Exe (mentioned in my article).

Photoshop 5.5 quality 6
        8     6     6     8    12    14    16    17 
        6     6     6     8    10    13    12    12 
        6     6     7     8    13    12    12    12 
        8     8     8    14    12    12    12    12 
       12    10    13    12    12    12    12    12 
       14    13    12    12    12    12    12    12 
       16    12    12    12    12    12    12    12 
       17    12    12    12    12    12    12    12 

Paint Shop Pro 7 quality 15:
        5     3     3     5     7    12    15    18 
        4     4     4     6     8    17    18    17 
        4     4     5     7    12    17    21    17 
        4     5     7     9    15    26    24    19 
        5     7    11    17    20    33    31    23 
        7    11    17    19    24    31    34    28 
       15    19    23    26    31    36    36    30 
       22    28    29    29    34    30    31    30 

Well, I don't edit and save as JPEG. I have a 10D now, and I shot raw and extract tiffs.

However, many many people would be intimidated by all of the steps I take.

Specifically, people like my parents. They have a computer, but they still think solitare is pretty cool, and really don't use it for much other than solitare, surfing the web occassionally, and now digital photography. And they wouldn't know how to install a CDRW drive. And they probably wouldn't want the hassles of burning and archiving the files on CD.

For people like my parents, the simplier the better.

Also, the only photoshop they would be likely to use would be if it came as a stripped down version with a digital P&S camera.