Improving image quality through algorithms?

[tarsands]

Hey Team,
I have a question for you, for an idea that came up at work today which is not really directly involved in photograph restoration but sort of is. Let me explain. Also if there is a better forum to ask this in please let me know.

I work for a seismic research company, and we basically take 3D photos of the earth more or less. We can orient our volumes of data in a way similar to a photo. That is, if I have a 3D volume ( a cube ) I can take slices of that cube at do processes on it independently, so it would be like working on a picture. My question is regarding what sorts of mathematical algorithms are used in photography that help to bring out details, remove noise and graininess, interpolate, and enhance edges. I am not really looking for software that does that, but maybe papers, or research that does this, as I believe I could eventually make it work for my field.

Again I know this is not what you usually get asked and if you can help or know of a better forum for me to cross post in, let me know.
Thanks kindly.

[chillin]

Re: Improving image quality through algorithms?

You could start here link

[klev]

Re: Improving image quality through algorithms?

3D scan data is not the same thing as a photo. You will have to be far more specific. As for algorithms, there are tons of them. It's too broad of a question. You can't just say "I want to improve this", because you'll get answers of comparable quality to the question. There's no way for me to even know what kind of data you have or your overall intention for it. There are resampling algorithms used to enhance the appearance of sharpness. They're implemented in most image editing programs, but you would need to take it into such a program. If you're dealing with shaders assigned to a cube, you would have to export UV maps under most circumstances. Right now I suspect you need to figure out what you actually have.

[tarsands]

Re: Improving image quality through algorithms?

3D scan data is not the same thing as a photo. You will have to be far more specific. As for algorithms, there are tons of them. It's too broad of a question. You can't just say "I want to improve this", because you'll get answers of comparable quality to the question. There's no way for me to even know what kind of data you have or your overall intention for it. There are resampling algorithms used to enhance the appearance of sharpness. They're implemented in most image editing programs, but you would need to take it into such a program. If you're dealing with shaders assigned to a cube, you would have to export UV maps under most circumstances. Right now I suspect you need to figure out what you actually have.

So for the purposes of what we do, the images we take are the same as a photo. They are completely independent of one another. Once we separate them out we don't need them to interact with each other, nor does the data in one image affect the next. My overall intention is to delineate edges of curvalinear lines or surfaces as may be the case, in very noisy environments. Basically i am looking for algorithm that use perhaps semblance or coherence to identify and enhance patterns that the human eye can't see but that are there.

The best research I have found so far on the topic is actually from the medical field. Imaging in CT scans and MRI's have already been pretty valuable in the ideas they use to delineate edges.

In the most basic sense, what I deal with is almost exactly like topographical maps of the subsurface that are quite noisy the closer you get to the surface. So I am looking for ways to identify topographic structures in noisy environments. We sample at incredibly fine rates and have all sorts of algorithms for dealing with interpolation vertically, once we get those patterns.

I hope thats more clear.

[klev]

Re: Improving image quality through algorithms?

I'm making the assumption that you are dealing with rasterized data, right? This still doesn't fully clear up exactly what you have available. When you say cube, I think you may mean a cube map. I am familiar with that. I didn't just look it up on wiki. It's just a way to employ environment mapping without the introduction of top and bottom poles or reconstructed topology, or maybe I'm still way off. I get that you are scanning light paths of either visible or non-visible rays, and the digitized versions of these lack the appearance of a fully continuous signal. Is the actual data raster data? Raster data would be in the form of square pixels as opposed to tessellated data. I get the impression that you have to deal with that. I'm not sure what applications you'll be using or the format of the data. If it's non-linear color managed data, it an be difficult. If it's fairly raw/linear data, you would probably need some combination of high and low pass resampling algorithms to diminish the appearance of noise, assuming the radial scale of the noise (for lack of a proper descriptive term on my part) differs enough from the scale of the signal data. Even then raster data is already resampled to fit in square pixels, so do keep that in mind. That's why rendering applications apply anti-aliasing to framebuffer samples prior to writing raster data.

By the way, radial was meant as a descriptive term, not a technical one. I was trying to think of a way to describe a maximum distance around a given pivot from any angle.