In this article, we show that the ideas behind the eigen-face technique can be easily extended to the OCR problem by using multiple image classes. In addition, we also describe a modification to the standard Euclidean distance measure used for recognition which gives very good results in this context.
In previous work, we demonstrated that a linearity condition was sufficient for us to estimate distortion from the output of the Synthetic Discriminant Function (SDF). However , many variants of the SDF have been developed. We extend the linear interpolating property to a class of SDF variants and use this result to analyse the properties of two of these… (More)