Pattern-Theoretic Foundations of Automatic Target Recognition in Clutter

Our goal is to apply Pattern Theory, as formulated by applied mathematician Ulf Grenander, to the problems of identifying targets in cluttered scenes. This effort leverages legacy work by the PI and others conducted as part of the Center for Imaging Science, founded by Prof. Michael Miller (originally at Washington University, currently at Johns Hopkins University).

Final Report

• As one big PDF file: (27 MB)
• Split into three parts: (part 1, 4.8 MB) (part 2, 7.4 MB) (part 3, 9.2 MB)
• Text files of the report body: (a file Microsoft Word seems happy with, but not TextMate) (a file TextMate seems happy with, but not Microsoft Word)

Presentations

• A Mathematical Theory of Automatic Target Recognition (atr_quantum.ppt, invited talk at the Workshop on Quantum Algorithms for Signal, Image, and Data Processing, UCSD, 7-9 Dec. 2004. It contains animations and is hence best viewed in "View Show" mode. (I also showed the Example Jump-Diffusion Process movie given below.)
• Target Recognition in Cluttered Infrared Scenes via Pattern Theoretic Representations and Jump-Diffusion Processes (ppt), presented at the AFOSR Program Review, Princeton Univ., June 2003. It contains animations and is hence best viewed in the "View Show" mode. The powerpoint file references the following movies, which are best played in an external player such as Quicktime that lets you manually step through frames:
  • Diffusion Example on AMCOM Data (mpeg)
  • Example Jump-Diffusion Process (mpeg)
  • Blob Estimation Example - M60 "Spreading" (quicktime)
  • Blob Estimation Example - M60 "Decaying" (quicktime)
  • Blob Estimation Example - Ship "Decaying" (quicktime)
  • Factory Example (mpeg)