How to optimize in image exploitation

The following image shows a frequency fingerprint of a radiating object, as seen by an infrared camera or sensor.

The dashed red line is the theoretical fingerprint based on a smooth blackbody radiation curve or model. This assumes that the object has certain properties which may or may not be actually present. These assumptions are the main reason why the actual observed fingerprint (the sold red line) does not match the theoretical model.

When data is taken of an object in space for example, all that we know is the frequency spectrum or fingerprint of the object. We usually do not know what the object is actually made of, it’s shape, etc. We have to make predictions or really educated guesses for that based on the observations.

Our goal at Pinkmoon is to develop a computer algorithm or model which can rapidly identify possible sources of discrepancies between the theory and observations, such as noise in the detector or camera, dynamics such as rotation, or background illumination.


One way to approach this problem is to do a kind of decomposition of the original curve into a set of known components (i.e. transformations or convolutions of the theoretical model). This is similar to a Fourier transform. Because multiple solutions might be possible, a cost function is desired which identifies which level sets or solutions might be more likely than others.

We are using open source tools such as Anaconda, Pyradi, and other technologies including AI to arrive at an algorithm for this problem set.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.