Tuesday, March 7, 2017

My semiconductor "event" pattern recognition patent

All I really wanted was to have my patent application US 20040175943 A1 get out there. I guess it did since I just found this on Google. The chi-squared formula applied by relative area, as in the above, is a key mechanism for measuring event position data against specific reference regions, such as illustrated:
Clearly the collection of fixed regions, with chi-squared calculated for each, gives you a mechanism for embedding patterns of dot scatter into a vector space - one dimension per region. Also you can see that there are families of region (the rows in 2nd pic) that differ by a group operation or "symmetry".

Here is part of the point, a program can be carried out with this way of measuring shape that goes further than the one I tried to do in grad school. So imagine representing a region by a pale gray transparent value inside the region. Imagine the darkness of the gray is a function of the chi-squared of the region versus a fixed scatter of event dots. Now superpose the transparency of many different regions over the pattern and I am confident the result is a sketch of the pattern. Thus in a real sense a pattern is the weighted sum of all the sub regions, weighted by the chi-squared. If I was a stronger mathematician, I would defend the formula

No comments:

Post a Comment