I am relatively new to serious digital image processing. However, I have jumped in with both feet. The comments below are some first impressions of PI.
I found LE a couple of months ago when I was searching for ways to reduce noise, specifically color noise, in photographs (regular photographs, not astrophotographs). I tried out a broad range of noise removal procedures, Photoshop plug-ins and software products, free, open source, and commercial. Neat-Image was the best of the lot, IMHO. After much time spent experimenting with the various settings for wavelets and SGBNR, I concluded that LE was more versatile and precise than Neat-Image.
Neat-Image has a great gui with some very nice input parameters for an unspecified set of algorithms that remove noise, preserve color, and allow camera background noise mapping, etc. "Algorithms," apparently, is all any plug-in/software is, including Photoshop. Image processing just is mathematical operations on the values defining each pixel in an image.
There is an endless supply of "probably useful" PS plugins, most of which cost $25 and up. One could spend a LOT of money buying "yet another plugin", each with its own "unique, wonderful, proprietary, best-for-purpose, etc, etc" - and totally opaque to the user - algorithm.
PI exposes the math to the user. I have learned tons about what image processing is all about, as I try to understand what PI is doing, and as I attempt to translate tutorials aimed at astrophotography to techniques that work with regular photographs. Having the math be more transparent, rather than completely hidden behind the gui like some kind of voodoo magic, is WONDERFUL!
For example, while cruising around the internet looking for information on tone-mapping and local contrast enhancement (beyond USM + masks in PS), I encountered the following claim made by Pixel Vistas (
http://www.pixelvistas.com/photolift/photolift.html) about one of their plug-ins, called "PhotoLift":
"PhotoLift takes advantage of recent advances in non-linear image filtering, enabling the aggressive amplification of local contrast in an image without modification of overall tonality or the creation of artifacts. The filter operates by separating the image into small-scale (detail) and large-scale (base) components.
"Unlike traditional methods of local contrast enhancement, strong edges in the original image are not represented in the detail component. This allows detail to be manipulated without producing unnatural halos or other artifacts in the vicinity of prominent edges. Furthermore, since the overall tonality of the image is largely contained in the base component, such detail manipulation does not significantly affect the large-scale tonal distribution of the image."
A little bell went off in my brain - PI can be used to separate images just as described above (if I understand P.V. and PI correctly, the terminology in the above quote from P.V. is reversed from what PI uses - "edges" (at least in regular photography) is small scale in PI, the "details" that P.V. talks about is what is left after using PI wavelets to extract the small scale "edges" portion of the image).
P.V. sells the plug-in, plus another plug-in that allows one to sharpen edges without touching the rest of the picture. Both plug-ins sound like guis covering a limited, single purpose, multiscale processing algorithm. In fact, each plug-in sounds like a limited version of one-half of "a trous wavelet" processing.
So yesterday and today I worked out how a procedure to use LE to do multiscale processing as suggested by the above quote from Pixel Vistas, extracting and cleaning the edges in wavelets and sending the balance (after a quick stop in SGBNR for noise control) to PS for local contrast enhancement, then back to LE to be recombined using PixelMath. AWESOME!!!!
Then I accidently stumbled upon the fact that PLC does Unsharp Mask. So I downloaded the latest PLC and poked around with Dependancy Walker until I found out why it wouldn't run on my computer (for some reason LE opens but immediately crashes on my computer unless I open it with Dependency Walker; PLC wouldn't even open on my computer without D.W., and when I opened it under D.W. it just exited with a fatal error before even bringing up the splash screen). It turns out that all I needed to do to get PLC to run was copy the dlls in the PLC Microsoft.VC80.CRT folder over to the WINNT/system32 folder, and now PLC runs on my computer without needing D.W.. Cool!
After I play around with PLC and LE multiscale processing a bit more (I can hardly wait to try out the HDR wavelets), I am going to download the P.V. plug-ins (fifteen day trial period) and experiment to see if there is anything P.V. can do, that PI can't do as well or better. I am just going to bet that PI will be the clear choice.
First comments on PLC:
NICE!!! user interface, as in quite attractive to look at. That was my very first reaction, when I finally got it started on my computer. After poking around a bit, it seems intuitive to use, too.
When I work in Photoshop, I use Pro Photo as my working color space. I didn't see any mention of Pro Photo in the PLC Color Management window. Nonetheless (and unlike most non-PS softwares that reside on my computer), PLC correctly displays and duplicates the image.
PI-LE also correctly displays a Pro Photo image created in PS and opened in LE. But LE is unable to duplicate a Pro Photo image without ***drastically*** altering the colors. There may be a color setting in LE that could fix this issue, but I couldn't find it and attempts to do so result in the program crashing (as stated above, LE doesn't like my computer - I have to run it via Dependency Walker).
Neither PLC nor LE can open a LAB-mode image, it seems. In PS, I work extensively in LAB, about as much in LAB as in RGB. There are things one can and should do with an image, using just the luminance, without messing with the color. And the LAB space is just more intuitive to me, makes more intuitive sense. But there are times when handling both luminance and color together, in RGB, is more aesthetically appealing. Fortunately within PI often one has a choice of where to do the processing. So I save my work in PS in LAB mode to RGB before sending it over to PI. But it would be nice to avoid repeated transformations.
PLC doesn't like tiffs with layers - it only shows the top layer.
One final comment. It is possible (and likely the norm for most users) to do image processing in PS (and most other image processing packages), without having any idea at all about what the image processing package is actually doing to the image. Most image processing packages just lay out a lot of "buttons" to push via the gui. Over time users acquires a repertoire of "button-pushing" via layers and masks and blends and filters and such, that can get the job done quite nicely. My own repertoire keeps increasing and I like the results I'm getting.
But at least for me, the whole "process of processing" via "button-pushing" with guis hiding "voodoo magic" doesn't really seem to get more "intuitive," more "transparent," so to speak, to the intended artistic goal.
I don't see any way that one can image process in PI without understanding, at least in broad strokes, how the available processes actually accomplish the intended goal. So far, for me, the process of learning how to use PI has made me feel quite "empowered" and capable of creating whatever artistic intention I can come up with. In other words, learning how the tools work is making the tools transparent to my purposes.
So, back to work, just thought I'd share my experiences so far with PI and regular photographs. Have you thought about starting a forum for regular photographs?
Elle
