The blues, revisited

I thought some more about the problem of the blue sensor becoming saturated in the case of wide angle shots featuring the sunburst, and came to the conclusion that it should be largely fixable in software, since we "know" what color the water is supposed to be.

I reasoned, that, when the blue component is near saturation, the green component

which has not saturated yet can be used as an indicator of what the blue component should have been. We can then change the pixel to one of the appropriate hue.

As we near the actual sunburst, the red component will start increasing. So the red component can be used as an idiciator of how much to shift the hue towards white.

So, I came up with a short program which does this. Its too clunky for use by anyone by me right now (I need to learn how to write PS plugins), but here are some preliminary results. All processing was completely automatic:

before:

after:

Once upon a time PS included filter factory. It is perfect for this kind of thing.

You can download it or one of its substitutes and write your filter as an interpreted plugin. When done, you can compile it and distribute it as a binary. Fixes the problem of having to do a plugin UI but it runs slowly.

I did my early channel mixing experiments with video this way. My filter determined what was lit and what was not by the red content and selectively corrected the backgrounds.

I want!

Cheers

James

Once upon a time PS included filter factory.

I don't know how comparable it is, but Photoshop Elements 2.0 has a Custom Filter function that lets you design filters. It looks fairly crude, and I've never tried it, but here's how the Help menu describes it:

Custom

Lets you design your own filter effect. With the Custom filter, you can change the brightness values of each pixel in the image according to a predefined mathematical operation known as convolution. Each pixel is reassigned a value based on the values of surrounding pixels.

You can save the custom filters you create and use them with other Photoshop images.

The filter factory plugin was similar but it allowed access to RGBA values. It also had YUV but that didn't work as I recall. The custom filter allows distortions and sharpening, etc, but it doesn't give access to channels if I understand it correctly. The filter factory plugin also allowed compilation.

Here's a link to a filter factory page. As I recall there are several sites like this one.

Chris, It's hard to tell on the tiny pic you posted, but it looks like your filter added some artifacts around the border of the fan and the water. Can you post a larger pic zoomed in around that area?

Yes. I think it did. I'll be trying it out on some real-res pictures. I don't personalyl have any high res ones showing the effect in a big way like this sample did. Probably the user will have to properly select the region to apply it to, but maybe I can make it smart about deciding what is water and what is not.

Rrefresh my orignal post - I have tried the program on a larger image which has even more severe cyan problems - Its saturated in both blue _and_ green. I'm not satisfied with the results yet.

Yo Chris G

Keep working on that program!

I did a lot of shooting. The sun and bright background shots were unacceptable but others taken with the sun far out of the frame were quite good.

I found that shooting in Aperture Priority mode provided me with a lot of latitude as long as I adjusted the aperture to keep the shutter speed between 1/60 and 1/125. Again, I seldom (like twice ever) used a SLR to shoot wide angle underwater, therefore I have a lot to relearn

Chris

I'm getting closer. Refresh the original post's pictures to see the current output.

The sample picture is a really vicious one to try to fix - it doesn't just have a blue channel problem - the green channel is saturated at 255 over a large area as well. My program performs pretty well on things that are just saturated in blue right now.

I noticed PS8 has a documented object model that can be automated with Javascript (as well as other languages). Adobe also documents objects that can be used to form a UI, so the clear implication is that you can produce the software described here (complete with UI) in Javascript and have it be cross-platform. Would be much easier that using the Adobe SDK and porting it to the other platform.

If you've installed PS8, there's a scripting PDF file you can read.

I haven't been able to find much information on the adobe javascript stuff. In particular,

I need to be able to do math on pixel values.

I haven't been able to find much information on the adobe javascript stuff. In particular,

I need to be able to do math on pixel values.

On PC's look in C:\Program Files\Adobe\Adobe Photoshop CS\Scripting Guide

There should be some PDF files there. The JavaScript PDF has got a lot of pages but I haven't read it. I see there is an example (page 118) that manipulates an image directly.

OK - noone seemed to agree with me on the bright blues of the D100 being much better than the S2, which I think goes green quickly. See this post referred to by Herb and Craig http://uwdigital.home.comcast.net/camera_noise/

Correct me if I'm wrong, but the bright end of the blue spectrum is uniquely noisy on the S2. I don't like camera wars, and I think the S2 produces overall less noise and more resolution than the D100, but this chart confirms that i'm not nuts. This green water is not Velvia-like in the least!

Good catch, Kasey. I see what you're referring to.

From what I see in the histograms, the S2 is a little green overall and balance moves a little toward yellow in highlights. I looked at the dpreview.com review of the S2 and noticed that the S2 had a consistently greener rendering in the posted samples. I'm curious why that's so.

I got my gretag color chart and grayscale sheet today. I can't reproduce the davebox test but I can do a similar comparision with my cameras.

I noticed PS8 has a documented object model that can be automated with Javascript (as well as other languages).

PS7 has a scripting update you can install as well. Haven't done it myself, but it exists, and can be scripted in Javascript, VB, and Applescript...

OK - noone seemed to agree with me on the bright blues of the D100 being much better than the S2, which I think goes green quickly. See this post referred to by Herb and Craig http://uwdigital.home.comcast.net/camera_noise/

 

Correct me if I'm wrong, but the bright end of the blue spectrum is uniquely noisy on the S2. I don't like camera wars, and I think the S2 produces overall less noise and more resolution than the D100, but this chart confirms that i'm not nuts. This green water is not Velvia-like in the least!

I'm not sure you can draw any conclusions about the bright blues between the two by looking at those histograms, since the D100 shots were slightly over exposed. I would say that it's clear that the S2 is noisier than the other DSLRs at the dark end.

Have a look at the images from which I got the histograms.

http://www.imaging-resource.com/PRODS/S2/FULLRES/S2DBAWB.HTM

http://www.imaging-resource.com/PRODS/ND10...S/D100DBAWB.HTM

http://www.imaging-resource.com/PRODS/EDR/...ES/EDRDBAWB.HTM

I can't say that I noticed the blues in the S2 being much worst than either the D100 or the Canon DR.

Those histograms are from the white->black graduated strip on the "Dave Box" with the R G and B channels pulled out.

If you want to look at blue water histograms, then I suggest looking at something blue, like blue water or a blue area on the Dave Box.

You can do that yourself using the same photos that Herb used, but looking at the L and M sections of the color chart and producing some histograms from that.

Cheers

James