Finding Edges in Noisy Scenes,
Abstract
Research into methods of identifying edges in a noisy scene has been an active field of investigation for many years. Treatment of the subject may be found in many books written over the past decade and many different approaches are proposed. Recently a survey and comparative analysis of the methods was made. The body of this paper is segmented into 4 parts. In the first we derive and define a 'Moment Operator' which we show to work well for step and ramp edges. Then, we define and characterize second order edges using the concept of the rotation of a point in a vector field and develop the detector analytically. In Section 3 we develop the algorithms for implementing the previously defined operators. Finally, in Section 4, these algorithms are evaluated using ROC curves and compared with previously known techniques. The detection of edges to isolate objects in a scene is motivated by many distinct problems. One such problem arises in a tracking system where the input video image is analyzed and the object to be tracked identified. Subsequent input and feedback to the drive controls causes the sensor to re-orient to a new position in an attempt to maintain the same x-y coordinate position for the object in the field of view. While this problem motivated the research that led to this paper, the results herein discussed are much broader in scope and application. The constraints imposed by this problem led to a method that is useful in high data throughput systems. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1980
- Accession Number
- ADA090416
Entities
People
- Alton L. Gilbert
- Raul Machuca