particle swarm enhancement query

[halsboss]

http://www.sciencedaily.com/releases/2008/02/080201093341.htm
ScienceDaily (Feb. 3, 2008) — A new approach to cleaning up digital photos and other images has been developed by researchers in the UK and Jordan. The method uses a computer algorithm known as a PSO (Particle Swarm Optimization) to intelligently boost contrast and detail in an image without distorting the underlying features.

anyone know of an avisynth filter which uses this approach ?

[chipzoller]

If the article was dated 3 February, 2008 it stands to reason the development is quite new and therefore probably no filters have been coded yet with this functionality.

[MfA]

There is no such filter, and there won't be in a hurry ... when they say it is "better than other approaches in terms of the computational time" they mean stuff like GA and SA (ie. also unusably slow for video).

[tritical]

From what I gather reading the linked article and abstract is that they are simply using PSO to optimize the parameters of some image enhancement technique. However, I couldn't find what the actual objective function being used to evaluate (rank) the modified images is (and that would be the interesting part). There is this sentence:


The objective of the algorithm is to maximize the total number of pixels in the edges, thus being able to visualize more details in the images.

but that doesn't tell you anything useful. It would be nice to have the actual paper, but I couldn't find it anywhere (except for a subscription site).

[foxyshadis]

I'm sure it's PSNR. It's always PSNR with academics, same way it's always gaussian blur or gaussian noise used to create the test images.

[tritical]

But PSNR compared to what? They don't have an original image to compare too. If I understood correctly, they have a video sequence and are trying to enhance it. This is done by modifying the parameters of some image enhancement technique on a frame by frame basis. The modified versions of an image (generated by varying these parameters) are ranked relative to each other. Therefore, the objective function must somehow rate the quality of these generated images, and that rating must be done without comparing to a pre-enhanced version. Otherwise, what is the point? You could simply return your pre-enhanced version.

[MfA]

They seem to be simply comparing their implementations of PSO and GA for optimizing a zero reference metric with constraints. The metric doesn't seem the main thrust of the paper, so it's unlikely to be new. No idea what it is, can't read the paper (http://inderscience.metapress.com/app/home/contribution.asp?referrer=parent&backto=issue,5,6;journal,1,2;linkingpublicationresults,1:120731,1).