Multistrategy Learning for Image Understanding.
Abstract
Current Image Understanding (IU) algorithms and systems lack the flexibility and robustness to successfully handle complex real-world situations. Robust 3-D object recognition, in real-world applications operating under changing environmental conditions, remains one of the important but elusive goals of IU research. We believe that an innovative combination of IU and Machine Learning (ML) techniques will lead to the advancement of the IU filed in general. IU itself has come to a certain state of maturity, in that we have today a good understanding of the essential components, their functionality, and the architectural issues involve. IU processes are commonly separated into three hierarchical layers, called the low, intermediate, and high level. At each of these levels. ML techniques can be employed selectively to improve the overall recognition performance. By introducing adaptation of task parameters; maintenance of internal representations and hypotheses pertaining to the observed reality: and learning new concepts and recognition strategies. The incorporation of learning into IU algorithms and systems will results in adaptation and robustness capability since learning provides automatic knowledge acquisition and continuous improvement of recognition system performance. (AN)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 15, 1995
- Accession Number
- ADA295440
Entities
People
- Bir Bhanu
Organizations
- University of California, Riverside