Working Papers in Acquisition of Knowledge for Image Understanding Research,
Abstract
Use of knowledge has facilitated complex problem solving in many areas of research. However, in the Image Understanding area, we do not have any systematic treatment and codification of knowledge that is useful in image perception. Further, we do not even have adequate tools for acquiring the necessary knowledge base. In this report we present an experimental paradigm for knowledge acquisition, discuss an analysis technique, and illustrate the different types of knowledge that seem to be useful in image understanding research. In the first paper, three major aspects of knowledge are presented: primitive Feature Extraction Operators, Rewriting Rules, and Flow of Control. A limited number of Feature Extraction Operators were repeatedly used by the subjects to specify location, size, shape, quantity, color, texture, and patterns, of various components found in scenes. Six types of Rewriting Rules were identified; assertions, negative assertions, context-free, conditional, generative, and analytical inferences. Flow of Control exhibited characteristics of an hypothesize and test paradigm capable of using imprecise, conflicting hypotheses in cooperation with others in a multi-dimensional problem space. The second paper discusses the picture-puzzle paradigm and the various ways in which it can be used as a tool for acquisition of knowledge. The third paper deals with a computer program that assists the transcription of typical protocols obtained from the picture puzzle tasks. Finally, the last paper of the report discusses the pros and cons of using eye-fixation data to acquire knowledge used in some of the tasks of the picture-puzzle paradigm.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 15, 1976
- Accession Number
- ADA118493
Entities
People
- Marty Schultz
- Omer Akin
- Raj Reddy
- Ronald Ohlander
Organizations
- Carnegie Mellon University