Deconvolution

[arith]

I was just looking at my wallpaper; an image I did a year ago and have even printed. Using Deconvolution I missed some artifacts that Topaz introduces. That is why I said it is hard to use; I've only just noticed the artifacts, but the more I practice the better I get.

Done about a year ago:





and my corrected version done without any acutance or contrast sharpening at all:





Obviously I don't like redoing images; but at least the print is reasonable and you have to know what your looking at to see the artifacts.

[Cantab]

A nice photo. I've downloaded a demo version of the deconvolution software by Quarktet but it's going to be a significant learning curve before I'll be able to see whether it will be of use for my purposes. They have what appear to be good tutorials to work through so that may help.

[John Morton]

I've been using a deconvolution plugin that comes with the Fovea Pro 4 set for Photoshop for some time. I have had great results when it works well but it is often hard to judge whether the settings I have chosen to use are doing what they should. The situations I use it in are extreme, though: to remove as much diffraction as I can from images taken with bellows mount macro lenses at f32 (so that I get a combination of maximum depth of focus and maximum sharpness). Here's an example that turned out not too badly:



There are different types of deconvolution; true deconvolution goes through a Fast Fourier Transform (FFT) into frequency space; the deconvolution is performed, and and inverse transform (iFFT) returns the image into a pixelated for that we can recognize.

The theory behind this states that all the information in an image is always still there so various defects of focus, lens optics, diffraction etc. can be corrected and the original image returned as what it should have been. I have heard that there are now camera systems which capture 'total images' in an unfocused form, which software can render repeatedly with any desired focal plane; so that is apparently a very real application of deconvolution as integral to imaging.

[Cantab]

John, an interesting photo. Have you come across the Quarktet programmes? I've downloaded a demo version of one of them but it looks like two of them (Tria and SeDDaRA) might be of interest: http://www.quarktet.com/SeDDaRA.html . I've read some of the science behind them but have no mathematical training so will have to take things slowly in trying to figure out how they work. The programme you've been using looks like it may be worth trying out.

[John Morton]

Hi, Bruce;

Yes that does look very interesting and I will have a look at it. I just had a look at the website for the software I mentioned, Fovea Pro 4, and it is still "under construction" as it has been for some time now.

http://reindeergraphics.com/index.html

One thing I highly recommend to every technically inclined photographer is a book by the father of the Fovea Pro software programmer (Chris Russ): "The Image Processing Handbook" by Dr. John C. Russ (5th edition, CRC Press, 2006, ISBN 0-8493-7254-2). The book is pretty expensive, but it is still a lot cheaper than the software!

Please note that there seem to be links on the Internet for downloads of "Fovea Pro 4.01" although the Reindeer Graphics website lists version 4 as the most recent. There are also links to a couple of companies advertising "uninstallers for Fovea Pro 4.01" so I am going to recommend caution to anyone interested in Fovea Pro 4.0 until I hear back from Chris Russ about exactly what is going on here.

[Cantab]

Yes, the book by Russ is expensive. I've discovered that a university library from which I can borrow has a copy of an early edition (1992) of Russ's book so I'll have a look at it and get an idea if it will be understandable by me.

[John Morton]

I've just heard back from Chris Russ at Reindeer Graphics who confirms that the "Fovea Pro 4.01 download" and "UNinstaller" is some kind of scam. He is going to look into it. So, if anyone is interested in Fovea Pro 4, make sure you purchase version 4 directly from Chris Russ at Reindeer Graphics, rather than another version from some other site. The purported version 4.01 that I happened upon appeared to be a download offered by CNET so whomever is behind this scam has gone to great lengths to make it appear to be legitimate.

[tclune]

One thing I highly recommend to every technically inclined photographer is a book by the father of the Fovea Pro software programmer (Chris Russ): "The Image Processing Handbook" by Dr. John C. Russ (5th edition, CRC Press, 2006, ISBN 0-8493-7254-2).

My field is medical image processing, and I must suggest that this is probably not a good choice for most photographers. Image processing as a field is geared to a very different thing than photography post-processing. The typical aims include tracking objects across multiple frames; finding objects within a frame; tracing edges in the presence of noise; etc. The point is that the goal is to perform technical analysis on images, either for machine control during manufacturing or for data extraction and analysis in such things as medical diagnoses. Maintaining or enhancing the look of an image is of essentially no interest in image processing as a field. There are many good books on post-processing that focus on using the tools of image processing for the purpose of enhancing the appearance of a photograph, and it makes much more sense for even a technically-inclined photographer to focus his attention on those books.

If you are dead set on learning something about image processing, John Russ' book is still pretty mediocre. I would suggest as an alternative any of the many texts called "Digital Image Processing" (Gonzalez & Woods, Pratt, or Castleman come readily to mind). One of my favorite -- although rather quirky -- image processing books is a delightful work by J. R. Parker, "Algorithms for Image Processing and Computer Vision." It is not a comprehensive overview of IP methods. Indeed, it appears to have been based on a course or two that he taught that were organized around particular IP projects. But it is both fun and sophisticated in the material that he covers. One of the things that is less than thrilling about reading a typical textbook is that the need for rigorous coverage of basic techniques makes them quite boring to read. Parker seems content to concentrate on particular methods that move him toward completing the task at hand. FWIW

[John Morton]

Agreed, image processing is a very technical application of photography; but, it certainly does put an emphasis upon science and this is something that I have found slowly dissipating from the field of photography as digital imaging replaces chemical based processes. I don't know if anyone is even discussing photography in terms of a balance between its artistic and scientific heritage these days, but I for one think that it is very important to keep the scientific element alive within photography.

A parallel example might be found within the martial arts, where I have been practicing Tai Chi for the last ~28 years or so. Once a very formidable system of self defense, taijiquan is now taught for the most part as a system of exercise designed to enhance the health of a practitioner. The self defense aspects are still there inherent in the artform, but those who understand how to defend oneself using Tai Chi are now few and far between. To me, photography without is backbone of science is like taijiquan without its self defense aspects: nice to look at but not as useful as it should be when the need to use its technical aspects arises.

That being said, I must of course agree that most people who are just gaining an introduction to photography would not find the technical field of image processing of much use or interest.

Still, you never know when that kind of information will prove useful. I use that material in work I do producing macro photographs of artifacts, and have adapted some of those techniques and processes for my general photography. It has even turned out to be useful in my daily life: some years ago, when I was working in a factory that produces pharmaceutical goods, the mechanics were having trouble with a subsystem on a new machine which places bottles of product into cartons. The difficulties they were experiencing were with the optical character verification system, which checked each carton for the correct lot number and expiration date. Since I was familiar with John Russ's text and had direct experience with his son Chris's software plug-ins, I was able to set up that system correctly for them and explain to them how it needed to do what it should be doing.

Since there were optical verification systems all over the plant checking for labels as well as lot numbers and expiry dates, I found my employment situation enhanced to the point were I survived two rounds of layoffs despite having been there for a far shorter time than most of the people who were let go when a major slowdown in production occurred. And, knowing how to successfully threshold printed material into clearly distinct blacks and whites so that optical scanners can successfully process it after picking it up, is knowledge that can also be used when producing masks in Photoshop.

My point here is, if you really enjoy photography it never hurts to expand your knowledge of the field because with digital photography, everything is changing and employment opportunities in photography are now very different than they were twenty years ago.

Thanks for the new references, Tom, I will most certainly look them up.

[tclune]

Hi, John. If you are doing work with machine vision, let me throw out one more book that I think is very useful, E.R. Davies' "Machine Vision: Theory, Algorithms, Practicalities." It does a very nice job of presenting different approaches to machine vision problems and indicating the strengths and weaknesse of each. Here is a link: http://www.amazon.com/Machine-Vision...machine+vision At heart, it is a round-up of recent and classic journal articles on many aspects of machine vision. It gives you the "Reader's Digest" version of a huge cross-section of research into important machine vision topics, ordered in a way that makes it a breeze to zero in on the techniques best suited to your particular machine vision problems. You can think of the book as the poor man's machine vision consultant. It's excellent.

[John Morton]

Thanks, Tom, I will certainly check that out. This area is still of interest to me, although I have moved on from that particular employment situation. The problems I faced there were fairly simple for someone well versed in photography; for instance, the cartonning machine they replaced had physically embossed lot numbers and expiry dates into the cardboard of the cartons using lugs of raised type. The new machine utilized an inkjet to print the lot numbers and expiry dates onto the cartons; and due to the length of time it takes to get print mats okayed in that industry (the new cartons had a blank space for the info to be printed within), the company was in a turn-around situation where they had to spray black type onto a proprietary blue back ground (blue being, of course, a mix of cyan and black ink). That's where I stepped in.

Other issues were much easier to resolve, such as small bottles of product on a wide conveyor belt being falsely rejected for bad labels due to the shallow focus of the optical character verification cameras. A simple increase in the depth of focus of the cameras through using a smaller aperture instantly fixed the problem and saved the company a pile of time and money being spent on manually sorting all the false rejects from the very few true rejects (none of which could ever be allowed to leave the factory - regulations are very stringent in that industry).