Design of Edge Detectors for Reduced Images.
Abstract
The development of algorithms to extract informational features from imagery is an area of active research. These algorithms enable computerized devices to automatically locate and identify objects in the field of view of a sensor. An important Air Force application is automatic target identification and weapon guidance. Edges in an image contain much of the information necessary to classify objects. This investigation has centered on finding methods for reducing an image so as to maximize the retention of edge information which was subsequently extracted. The Hotelling transform which reduces image data so as to minimize intensity mean-square error (IMSE) in the reconstructed image was also found to have significantly better edge retaining ability than simple averaging. The reconstructed edges were quantitatively compared to those in the original images using MSE and receiver operating characteristic based measures. One such measure used was the gradient mean-square error (GMSE). Both the reconstructed IMSE and GMSE using the Hotelling transform tend to decrease as the encoding block size increases. An equation relating GMSE to IMSE was developed. For image gradient blocks that are independently reconstructed, the linear transformation matrix A that minimizes the reconstructed GMSE and in that sense maximizes edge retention was derived. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 1984
- Accession Number
- ADA145700
Entities
People
- D. J. Healy
Organizations
- Georgia Tech