AstrOmatic forum

Hi there!

I have been developing a way to use EyE to enhance the way we are currently using SExtractor to find objects in our fields. (The general use of it I guess). At the moment I have not had very much success with getting the EyE to recognise what is in the model images. The patterns it picks out don't seem to have any correlation to the image at all. i.e. A model image may have 2 or 3 gaussians on it but EyE will pick out over 100 patterns. Have I misunderstood what the patterns are? Has anyone come across this before? I know that there must be something with the model images but I am unsure as to how they are so wrong….

Any insight would be very much appreciated!

Cheers,
Elise

Hi Elise,

How large are the objects? Eye, by construction, is only able to "identify" very small patterns (the size of the "retina", basically), such as image glitches and, in some instances, point sources. Now of course one could just increase the area of the retina to, say, hundreds of pixel; but this won't work well in practice because of limitations in the training algorithm.
A better approach for working with sizeable sources would probably be to do some dimension reduction of input images and replace the pixel basis with a carefully chosen set of broader functions (with sparsity properties, similar to the receptive fields found in the visual system).
Cheers
- Emmanuel.

Hi!

The objects wouldn't be more than say 20 pixels across at the largest. Its a 120 arcmin field of view. What I am hopping to do is get EyE to be able to learn the patterns of stars and roids. I will have a look at how increasing the area of the retina goes. If that doesn't work I will definitely have a look at dimension reduction.

Thanks for your help,

Elise