I have an older version of PS, CS2. I recently came across this image, which I would like to "convert" back to the original, if possible. Any suggestions as to how to accomplish this? I think it's bas relief.
The short, correct answer (as given above) is, "no". However, one can kinda-sorta somewhat soften the bas relief effect by simply applying a series of Gaussian Blurs to the image, interspersing between each a "levels" adjustment layer to spread apart the histogram just squished by each of the Gaussian blur operations.
There is some actual image processing theory behind this. The bas relief effect is generated by something close to a Laplacian operator acting on the original image. This operator looks at the difference between a central pixel and all the pixels around it out to some range. This suppresses the spatial low frequency components of the image and accentuates high frequency spatial components in the image to give the very flattened effect that the OP posted. Running the bas relief version through a series of spatial low pass filters (eg, a Gaussian blur) is, generally speaking, the inverse of this process and adds a bit of depth back to the image, assuming the gods are smiling on you. ;-)
Unfortunately, there are many, many complications that prevent one from performing a good inversion, not the least of which is that we don't know exactly what algorithm was used to generate the bas relief version. That being said, at least the image can be sorta moved in the right direction with some simple tools native to PS.
that is high pass filter and like others have said no it cant be changed back.
an image doesnt store what it use to look like so you only have what it is now
Dataflow, presumably you have taken freshman calculus? If so, then you should know that you can take the derivative of a function, integrate the result, and get the original function back. This is closely analogous to what happens when you take an edge enhancement spatial filter (ie, a derivative-like operation - Google "Laplacian filter") and then blur the result (ie, an integration-like operation).
By your reasoning, the fact that "...a derivative doesn't store what it use to look like.." would imply that integration shouldn't work, whereas, of course, it does.
As I said earlier, in the real world, reconstruction is difficult and usually only partially successful for the reasons I gave in my 1st post in this thread (and others).