A class of linear maps for error corrective dimensionality reduction of binary templates

摘要

Binary templates are optimally encoded in a reduced dimension by a proposed class of linear maps that preserves the local neighbourhood and a prescribed minimum distance between the prototypes to a workable extent. This in effect generates a nearness criterion suitable for template matching with a level of error correcting capability in the reduced space while requiring only a fraction of memory storage space and boolean operations that would have been required otherwise. Characters and symbols may now be designed with reference to separation and shape but with a comparative freedom from the constraint of dimension, while a volume of such data can be transmitted with the speed and bandwidth of the encoded data. Some of the principal problems associated with template matching are thus overcome. Here all the necessary operations are performed in the finite field GF(2) and the methodology developed can be implemented in a microcomputer with improved system performance and economy.