Detection and Identification of Objects Embedded in Cluttered Fields: A Reconnaissance Problem.
Abstract
The Air Force is concerned with the complex problems of identifying objects of interest in aerial photographs. This work is an expansion of a mathematical model of the human visual system into the realm of cluttered fields. The study considers reported modulation transfer functions (MTF) and demonstrates that the dissimilar curves can be explained by a task-related MTF. A series of aerial reconnaissance photographs are used as input to a computer-controlled flying-spot scanner. The photographs were scanned with a coarse 32 x 32 grid. The algorithm then determined those regions which contained prospective objects of interest by use of adaptive thresholds. The algorithm located 88% of the targets. A clustering algorithm was developed for the purpose of defining useful prototypes. The prototypes eliminated more than 75% of the false alarms, with no additional loss of targets, or 89% with a recognition of 82% of the detected targets. In spite of the false alarms, the process only requires 2 to 8% of the picture elements required for a television system transmission of each photo in a television frame. The 8% is for transmission of all targets and false alarms including redundant targets and false alarms obtained by the detection algorithm at several threshold levels.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1975
- Accession Number
- ADA017187
Entities
People
- Roger A. Gagnon
Organizations
- Air Force Institute of Technology