Machine Vision Through Machine Learning.
Abstract
This research has been concerned with the development of initial methodologies and vision systems capable of learning descriptions of visual objects or scenes, and the application of the learned descriptions to the efficient recognition of objects in a scene. The underlying motivation for this project is that learning capabilities will make computer vision systems adaptable to a wider range of practical problems than current vision systems that in most cases lack leaning capabilities. In this project, we concentrated on the following topics: (1) Development of the MLT ('multilevel logical templates') methodology for learning image transformations that characterize classes of visual objects. (2) Implementation of the MLT methodology and its application to the acquisition of texture descriptions by learning them from object samples presented in a scene under varied perceptual conditions and noise. (3) Development of methods that use a simple form of analogy for learning visual concepts (the PRAX project). (5) Application of the developed methods and systems to selected practical problems in the area natural object recognition, object detection in a scene, and target recognition.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 15, 1995
- Accession Number
- ADA307591
Entities
People
- Ryszard Michalski
Organizations
- George Mason University