A Determination of a Sufficient Statistics in the Problem of the Recognition of Images Subjected to Admissible Transformations,
Abstract
The report discusses the solution of the problem of determining the sufficient statistics for classifying optical images in pattern recognition, using the concept of permissible transformations. The idea of permissible transformations consists of the fact that a set of images corresponding to one pattern in the absence of noise can be represented as specific transformations of a given standard reference image. The correlation recognition algorithm is based on this concept; it calculates a set of correlation functions and its optimality is determined with the aid of sufficient statistics. The new statistics is called sufficient statistics. The present study shows that the conditional probability density function under specified conditions depends only on two factors (1) the most probable value of the optical transformation parameter, and (2) the square of the distance between the initial image and the reference image. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 17, 1971
- Accession Number
- AD0731911
Entities
People
- L. Svyatogor
Organizations
- National Air and Space Intelligence Center