Context classifier
Abstract
A pixel can have certain properties when viewed in isolation, which change when viewed in the entire context of the image. A theory, and an algorithm are presented for a context classifier which gives each pixel the highest probability label, given some substantially sized context neighboring the pixel. The algorithm takes the form of a recursive neighborhood operator, first applied in a top-down scan of the image, and then in a bottom-up scan of the image. Applied to a simulated image generated from a real Landsat image, the context classifier showed better accuracy than the Bayes classifier. On real images, the contex classifier showed a 4 to 8 percent increase in overall classification accuracy over the context-free Bayes classsifier under a Gaussian-distribution assumption.
- Publication:
-
IEEE Transactions on Geoscience and Remote Sensing
- Pub Date:
- 1985
- Bibcode:
- 1985ITGRS...1..247H
- Keywords:
-
- Classifications;
- Image Analysis;
- Photointerpretation;
- Remote Sensing;
- Accuracy;
- Algorithms;
- Classifiers;
- Dynamic Programming;
- Probability Theory;
- Satellite Imagery