Image processing and image identification

[Bryan]

Hi:

A new guy with a basic question.

When I look at posted images I see what appears to be a central dark band of significant width, then an 'illuminated' band. I presume that the illuminated area is all that is useful, and that the change in dynamic range with range in the illuminated band has been corrected for. Correct so far?

In many images I see a chaotic image of the bottom, with often no man-made objects in sight. I also read that SSS has been used to characterize the bottom - I believe automatically - probably on the basis of texture and structure.

Does anyone use image processing to identify man-made structures from the bottom - on a personal level - ie not mine detection guys?

My background is in image processing of sea ice from radar images - and the problem is similar as far as classification is concerned.

Bryan

[ChrisPreece]

Target recognition/Image processing

Bryan,

I think a lot of people on this website are doing their own image processing 'by eye' more so than anything else.....I think so anyway.....if anybody is using 'target' recognition software I can't think they've mentioned it. It's more the case that people know what they're looking for....essentially things that don't 'fit'....straight lines, right angles and the like...the assumption being that things that don't fir are man-made.

[Bryan]

natural and man-made objects

Chris:

Thank you for the info.

Just for everyone's edification/dismay natural objects tend to be 'fractal' which means they look about the same at different sizes. Man-made objects as you point out tend to be linear or have smoothly changing curvature and often have orthogonal (grid) structures imbedded in them.

A natural fractal process is equivalent to chaos which is usually measured by 'entropy' which in turn is the building block of information theory. More disorder -> more information (removed).

Information is made up of textural and structural information. A natural process embedded in other natural processes is identified by a relationship between them.

Also the connectivity for objects having this relationship is unique.

So when a man-made object comes along with different texture/structure and connectivity it can be 'detected' - well at least within the accuracy of the method and quality of the image.

I need to add that there are many other conceivable methods that can be used. The above way is what I happen to be most familiar with.

Accordingly if anyone hears of what is being used, please post it here.

Thanks

Bryan