CHANGING ACTUAL PIXELS??

[Bill McCrystyn]

This is a quote from a new PS book that I read. NO Savvy?

“Remember that with raw files almost all the settings we specify in ACR are actually changes to the algorithms used to convert the information captured by the photo sites on the sensors into a visible image, rather than changes to the pixels. Although we work as non-destructively as possible in PS, ultimately all the changes we make in PS modify the pixels. The greater the changes we make there, the greater the potential for some image degradation. Changes made in the raw converter do not generally result in image degradation—except when clipping is introduced or areas are cropped out. However, the changes made in ACR to Jpegs and TIFFs are changes to the actual pixels—which is why we recommend shooting in RAW for most nature photographers.<O:p</O:p

[Alfred Forns]

Nothing new there Bill !!! All seems well !!!

Remember with JPEG you are literally throwing out a number of pixels then making he up when opening. Something fun to do is save a jpeg ten times then compare to the original !!!

[Bill McCrystyn]

Yes, I compare a L JPEG original with an original uncompressed 36 mb TIF out of the box and it's a joke but that's not what they are saying here. They are saying "change", not getting rid of. And how do you change a JPEG or TIF with/in ACR?? And what kind of degradation and to what extent?? This whole paragraph raises more questions than it explains. Can anybody actually explain what there talking about?

[Axel Hildebrandt]

I think the author tries to explain the difference between destructive and non-destructive formats and ways to postprocess. The phrasing could be clearer. I'm not sure whether he wants to say that TIFF is like JPG in that respect.

[Roger Clark]

Hi Bill,
This can be a tough one to explain and the actual differences are subtle.

First, a few things about photoshop. While a great photo program, it is not strictly scientific. For example. I recently had the experience of reading a file to check some values for a science project, only to find the values different than I expected. I was pretty sure the values were correct, so I wrote a test program that created images with values I knew. Photoshop not only gave a different answer, but if photoshop wrote out the values to a new file (I'm not talking compression effect here), the values were different. Photoshop changed values even when I did nothing but read a file.

Ignoring that problem, whenever you read a file (tif, jpeg, png, etc) and change something, like curves, you are changing the values of the pixels in the file.

When you read a raw file in ACR and apply curves the effect is still to change pixels in the output image (and ultimately the file your write out), the curves or other changes you apply are changes to the color filter array algorithms that interpolate between the red, green, blue sensor pixels. The raw converter must do some quite sophisticated math to interpolate color between the sensor pixels, and do a good job in the many different conditions posed by photography. Reportedly many raw converters use similar (or even the same) algorithms used in DCRAW, a free program with source code (in C). If you are curious, try dcraw and look at some of the options for conversion. There are some iterative methods that take some compute time but produce pretty good results.

But back to the question, it's a subtle difference of "changing actual pixels" or not as both have the effect of changing the output. If the input data have the precision (16-bit tiffs should for all current camera images) then it really doesn't matter much. A greater effect would be the actual algorithm for raw conversion. Given a good basic conversion to 16-bit files, then changing curves would make no noticeable difference in my opinion. Eight bit, however is barely enough for some images, and when you start bringing up shadows, or showing detail in highlights, the bit posterization can begin to show.

Early in DSLR years (pre 2006), I did not like most raw converters. I used the ImagesPlus converter (astronomy software) to 16 bit, then did my work from there, some in ImagesPlus and some in photoshop. Photoshop, by the way doesn't do actual 16-bit math (at least versions I've worked with). Internal to photoshop images are treated as 16-bit signed integers, 0 to 32767, rather than the tiff specification of 16-bit unsigned images 0 to 65,535. So photoshop is losing one bit of precision, and also explains way on reading a 16-bit tif, the values must be changed to fit the data into a smaller space. The problem I first describes above was with 8-bit files. There is no excuse for changing 8-bit data in my opinion.

If you stack images to boost signal-to-noise ratio, you can quickly get beyond photoshop's 15-bit precision. When doing that, I use other image processing systems, and work in floating point or 32-bit integers.

I do not know if this helps the original question. I think the description in the book is too subtle to be concerned with, as both change the output pixels and that is what we are looking at.

Roger

[Bill McCrystyn]

Thank you Roger. Now it is more apparent what they were saying. You also have given me the impitus to look into changing to Raw 12 bit. Although often subtle, the combined effects in post production of noise, color bandwith and perhaps most visable, posterization, are indeed worth looking at.