The Performance of Edge Operators on Images with Texture.
Abstract
This report evaluates the performance of several edge operators on real world images with texture. Two types of edge detection are shown to be important for the analysis of real world scenes. First is the detection of major surface boundaries (called macro edges), and second is the detection of the surface texture element boundaries (called microstructure edges). Six edge operators are then evaluated to determine their performance at both macro and microstructure edge detection. The image data set for the evaluation consists of one 'blocks world' image and five difficult real world scenes. The Hueckel operator is shown to be the most thorough and sensitive to low contrast edges, and the Kirsch operator is shown to be the best for fast, conservative, first evaluation. A distinction is made between sparse and dense microstructure to explain the apparent low contrast in low resolution microstructure. Finally, a strategy using goal-guided search and planning is mentioned as a way to take efficient advantage of the best properties of both the Hueckel and Kirsch operators.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1974
- Accession Number
- ADA006230
Entities
People
- Bruce Bullock
Organizations
- HRL Laboratories