The Measure Objects Module

Measure Object Properties

The Measure Objects module calculates parameters for objects, and classifies them based on their values.

Each OBJC is passed to the Measure Objects module individually, so all passbands can be measured simultaneously.

Actually measuring the object's properties is simple enough to describe, although some of the algorithms are rather complex.

The first thing to check is if the object's had its wings subtracted; if so we assert that it's already been measured (presumably as a bright object) and return. We should not remeasure any bright objects, as they have had power subtracted from their wings

Then we find a good centre for each band, a step that's only important if the object being measured is the product of the deblender. Then a canonical centre is found. If the object is detected in the band denoted mo_fiber_color in the frames parameter file (usually r'), the centre in that band is the canonical centre. If it isn't detected in that band, the band with the largest peak counts is used. It would be better to use e.g. a fibre magnitude, but one is not available at this point.

Once the canonical centre's in hand, we process the object in that band, followed by the other bands in no particular order. For each band, we:

If the object is blended, the deblender is run and this entire procedure is repeated for each child.

Classification using a Decision Tree

Having finished measurement, we are now ready to classify. Classification is performed using a decision tree, which is stored internally to PHOTO as a TREENODE structure (with branches connecting it to other TREENODEs). Some human must have created a decision tree appropriate for the data, and placed it (currently, in the form of an ASCII file) into the PHOTODATA directory, before the pipeline is run. It will not be possible to modify the tree for each frame, or even for each run --- we will probably end up using a single tree for at least an entire season, if not for the entire survey.

As of May, 1995, we have the ability to create decision trees relatively quickly. However, we have not spent time to try to create trees which yield high accuracy; that will be done later, when we have real test data.