Findfeature multi images

A transfer from the original Astrodiscuss forum (from drschenk)

trying to use findfeature to increase control network density, but I dont understand the “algorithm” coding (eg., where do I find example?): how to improve the quality of points selected, reduce their number and increase the number of measures per point (ie., the number of images used per point).

the basic problem is that I get several thousand points but few are really robust or useful, and overlapping images are rarely used or erratically.

transferred (from lwellerastro)

Hi @drschenk, you have a number of topics in your discussion that may be making it hard for some to reply to.

I’ll take a stab at it but this is really a very large topic.

See Example 2 via the very detailed program documentation for all available algorithms and their parameters: USGS: ISIS findfeatures Application Documentation
I refer to this document frequently and find something new every time. It’s loaded with information, but it may be overwhelming.

I don’t really have an answer for this since I’m not exactly sure what you mean. Fndfeatures will find matches within a few pixels of the mark. Some may be at the subpixel level, but many won’t. You will need to run pointreg on your network to get subpixel registration. But you would only do so after you have run cnetcombinept (see below).

You might try different algorithms and parameter settings if you haven’t done so already. I have used findfeatures for a number of data sets now (Galileo Europa, ISS Titan, Kaguya TC Moon, currently refining things for HRSC Phobos) and how found fast/brief to be the most successful. I used that along with sift/sift with tweaks to the octave parameter for Galileo along with fast/brief because it seemed to fill in and find points where fast/brief did not. I honestly don’t have a great feel for the general algorithms or what the settings should be and what to use where so lots of trial and error on my part. I keep using what works which has mostly been fast/brief.

The program cnetcombinept will help reduce the number of points in the network as well as increase the density of points by combining points based on user input. A larger imagetol (pixel distance for combining points) requires a larger search/pattern deffile for pointreg. Measures within the tolerance in point B that are combined into Point A have a different location and will be offset, so I try to keep imagetol lowish (3-7) depending on the data set and how successful I think I will be in coming up with a pointreg deffile to capture the offsets.

The program cnetthinner will reduce the number of points in an image based on user input as well. I use this program very frequently and after running cnetcombinept and usually after pointreg, but have also run it before pointreg for very, very large networks.

You might also have a look at this bug post on github: Findfeatures - "Can't invert empty matrix" Errors · Issue #4639 · DOI-USGS/ISIS3 · GitHub

Not so much for the particular error, but Kris Becker and I have a detailed conversation about findfeatures and overall procedures and he has an awesome summary of things he has made standard practice for his processing. I do much of the same and it might help you some.