What are non-destructive edits and do they exist?

Question

This is going to be somewhat provocative and more about terms and definitions, so if you feel it should be closed or moved to meta, vote so.

To the question - there is a lot of talk about non-destructive edits that can be done during post-processing, but I'm a bit puzzled by the term non-destructive.

Edits to RAW file are said to be lossless/non-destructive, because original data is kept. But how can I call an edit non-destructive if data is lost in the result? Same can be said about JPEG - if I keep the original data, my edits are non-destructive because I can always go back to original (JPEG is a lossy format, but that's different story, we can as well talk about TIFF).

So there is a difference between lossless/non-destructive edit and lossless format, ie changing RAW to JPEG itself is destructive edit, because we lose information as the result. But we also might lose information in the result if we do WB adjustment to RAW file (the photo is bluish, we add more red and clip the red channel in the process).

Is the WB edit for example non-destructive in the meaning that we couldn't do any better in the same conditions originally? But what if we didn't take WB edit as non-destructive in the first place and used filters to get better data?

So what are the definitions of non-destructive and destructive edits?

Answer 1

Like Alan said, usually when discussing edits, we refer to "destructive vs. non-destructive", where as when discussing file formats, we refer to "lossy vs. lossless".

If we take white balance, for example. Is adjusting white balance truly "destructive" when performed on a raw file? A raw file can not be viewed in its native state, as its just a bunch of "raw" sensor information and camera data. The information from a bayer array is not viewable directly...it has to be processed to create RGB pixels that may then be viewed as an image.

If I render a RAW file with all default settings, am I "destroying" information? Or am I simply interpreting it? When I change the white-balance setting, am I "destroying" information, or simply changing how I interpret the information that exists?

Lets contrast that with JPEG. A JPEG image starts out "lossy", since the original data has already been interpreted. Any adjustments from this point on are not from "original" data, they are from a previous interpretation. Technically speaking, within considerable limits, you could make non-destructive edits...but only within considerable limits. The data has already been partially destroyed, so you have to wonder, how much more destruction am I causing by adjusting poor data?

When I work with a RAW image in Lightroom, and adjust exposure, white balance, tone curves, etc., all of those edits are applied to original data. Every additional adjustment is refactored into the RAW processing, and applied to original data. If I adjust white balance several times, I'm not losing any data...the "total" WB adjustment is applied to the original data when it is rendered on the screen. Its not as if one WB adjustment is applied, then another on top of that, and another on top of that, which indeed would be destructive. If you apply a tremendous amount of edits to an image in Lightroom, you'll start to notice lag as you zoom or pan the image, make additional adjustments, etc. This is because any adjustments are reapplied to the original RAW data when it is rendered on the screen.

Assuming you do "clip" reds in a particular WB edit. The information is not actually clipped, since that is just part of the overall processing applied to the raw bayer sensor pixel data when it is rendered to the screen. If you make another WB edit later on to "recover" those clipped reds...they can be recovered in their entirity, without any loss, since its just a step in a processing "pipeline" that is executed any time the RAW image is updated on the screen (i.e. zooming in, panning, making other edits, etc.) The only time information is actually lost when editing in RAW is when you save to a normal image format. It doesn't necessarily have to be JPEG, you will still incur a loss when saving to DNG or TIFF.

Answer 2

The terminology is usually called "Destructive vs Non Destructive" editing.

The idea is that with destructive editing, information is lost. The information that is lost isn't truly lost, you can typically issue an "undo" which will revert the change, but this is not always guaranteed.

With non-destructive editing, edits and the original are kept separate, and only applied upon final save. These edits can be reordered at anytime, and in an arbitrary order to produce varying results.

As che said, Photoshop Adjustment Layers are non-destructive.

Answer 3

I think that lossless edits keep the original file together with change data.

So if you have a raw file and apply some color correction curves, and then some other color correction, you still have original data and two corrections, which can be computed at arbitrary precision, when you need to actually print the image.

AFAIK Photoshop adjustment layers work this way.

Answer 4

A RAW photograph will always be non-destructively edited, that is until you produce a JPEG or other system usable format that you can then either destructively or non-destructively edit further.

The key is in the software you use. Lightroom is, by definition, a non-destructive editor. Photoshop on the other hand is not, unless specifically used in a way which does not change the original image layer.

Lightroom stores the photo's settings and adjustments in it's database while the RAW file (and it's original settings saved by the camera) remains intact and untouched.

The edits are only destructive applied to the exported image, and even then the RAW file still remains the same. The good thing about Lightroom is it takes this approach to all files, JPEGs, TIFFs etc.

Answer 5

There is also another type of non-destructive editing (we should call it fake-non-destructive). digikam introduced non-destructive editing for all type of operations, even cropping/resizing etc. But I called it fake since it just saves a copy of the image before you started editing it.

You could do the same for your self with any other photo editing software by a simple backup before you start your work. digikam just utilized this (maybe) best practice and wrapped some tools around to make it more convenient.

Answer 6

Editing is making changes to something, which makes it a destructive process by definition. Non-destructive processing treats the file containing the original image as sacrosanct and does nothing to modify it.

Say I import an image to a program and take these steps to get it into a state where it can be released to my customer:

1. Reorient the image vertically.
2. Crop to 1:1.25 starting at some point (x, y) and having dimensions of width x height.
3. Adjust the exposure by +0.63 EV.

With destructive editing, saving the image at this point would permanently lose the pixels I cropped away in the second step and any colors clipped to white by the third. If I come back the next day and decide I'd overdone the exposure adjustment, the only option I have is to darken the entire image. The parts that were clipped just become shades of gray instead of what they were. The program will have replaced the original image with the processed version and leave no trace of the original or what I did to get it into its current state. Needless to say, none of that is recoverable.

Non-destructive processing isn't editing in the traditional sense. It's more like the markup done by proofreaders in that it shows what should be done to improve the image while leaving the original intact. Instead of overwriting the original image with my changes, non-destructive processing programs keep a separate list of what was done and use that to create what's displayed on the screen or produce the final output. This means that when I want to change the exposure adjustment made in step 3 above, I simply remove that step and the program shows me an image that's only been reoriented and cropped. Then I add a new adjustment to the list and the program goes through all three steps to show me the new image.

The change I made here was destructive edit to the list of changes but left the original image unmodified.

Answer 7

Lossless / nondestructive edits are reversible and would entail transforming and/or appending the data in a way that could be reversed.

Two easy examples of lossless edits are:

1) rotation by 90 degrees

2) quadrupling the resolution by transforming each pixel into an identical 2x2 quad pixel (would work for some image formats, probably not JPEG because of the way it stores information)