Stochastic Systems Group
Home Research Group Members Programs  
Demos Calendar Publications Mission Statement Alumni


Anomaly Detection and Localization, A. Frakt

Tomographic data are acquired by strip integrals over the image domain. The strips are organized into parallel sets at different angles. Here we show sets of strips at two angles.

This is an image domain view with a random background field and an anomaly superimposed in the upper left corner.

This is a view of the tomographic data (also called a sinogram) corresponding to the image domain shown previously. Each column represents the data collected at a particular angle. The anomaly corresponds to the black sinusoidal swath running through this image.

Our approach to anomaly detection and localization consists of a sequence of composite hypothesis tests. The anomaly (corresponding to one hypothesis) is successively localized by eliminating groups of hypotheses. In this figure, the shaded composite hypotheses are retained while the others are discarded. For more details please see the publications linked to my homepage.

Problems with this site should be emailed to