Joy, thanks for sending me your files to work with.
As I speculated, your problem is caused -- more precisely "revealed" -- by how Photoshop displays images at less than 100% resolution.
Here are a couple of screenshots that illustrate.
I'm displaying the same image at three different resolutions, and I've added a white rectangle so that we can tell exactly what's being displayed where.
First, here is the .psd file that you sent to me, displayed by Photoshop in its original layered form.
Then, here is the same image after flattening.
The differences will be more clear if I show just the image portions in side-by-side format -- on the left before flattening, on the right after flattening:
As you can see, there's no difference in the actual pixels, which are shown in the 100% and 300% views.
The difference is only in the reduced view, the one shown here at 33.96%.
Clearly, Photoshop is generating that reduced view differently in the two cases. I don't know exactly they do it, but obviously the method is very sensitive to the scale that is used.
Consider the following animation, which shows the very same image displayed at 33.96% and 35% :
Now, if the resizing for display were being done perfectly, all that we would see in the animation is the image alternating between two slightly different sizes.
In the gestalt, I think we can see the size change happening.
But concentrating on say those vertical stripes at the top of the image, we see two very different patterns -- and yet all that's been done is to display the same image at two slightly different scales!
If you look carefully at the rest of the pixels inside the white box, you'll see the same thing going on. These two screen displays are wildly different from each other, and yet all that has changed is the display size.
I have not tried to reproduce your problem with Camera Raw, but I have no doubt that it's due to the same underlying problem. ACR is probably showing you a reduced size image corresponding to the layered form, but when you send the image to Photoshop it's generating a reduced size image corresponding to the flattened form. I'll bet that if you compare before and after Open Image at 100% instead of size to fit, you'll see that the colors are nicely preserved between actual pixels in ACR and actual pixels in Photoshop.
Now, at the top of this post I wrote that your problem was
caused, more precisely "revealed", by this issue of how Photoshop generates reduced size images for screen displays.
The reason I say "revealed" is that, given the actual pixels you're generating, you're going to have the same problem with any display method.
The image you sent me is size 3908x2602. If you had a monitor that large, so that you could display the whole thing at actual pixels, then when you stood far enough back that your eyes could not resolve individual pixels, what you'd see would look pretty much like what Photoshop shows at reduced scale for the flattened image. In a word, "blah".
Likewise, if you tried printing the image at full resolution, you would get "blah" -- probably different blah from what the monitor shows, because of ink spreading, but blah nonetheless.
Looking at it another way, those beautiful colors and patterns that you so enjoy are not generated entirely by your exposure/contrast/levels adjustments. Instead, they are generated by those adjustments, followed by some unknown process that ends up computing or selecting intensely colored components for display at reduced resolution.
Indeed, doing a screen capture of the exciting colors may be the only way to preserve them!
I hope this is helpful. It's an interesting problem. Obviously this sort of thing doesn't happen with typical images, or Photoshop would be using a different method. What reveals the issue is your use of the low level wavelet data, which by construction packs most of its information into variation between neighboring pixels. Given my mathematical background, I'm actually curious why there's any large scale structure in this image at all. I think the multiscale decompositions that I'm most familiar with would not do that. But that's a matter to be explored at another time, because I suspect it will take quite a while to make sense of.
--Rik
