Abstract In order for neural network technology to make useful determinations of the identity of letters and numbers that are processed in real time at a postal service sorting center, it is necessary for the neural network to "learn" to recognize accurately the many shapes and sizes in which each letter or number are formed on the address surface of the envelope by postal service users. It has been realized that accuracy in the recognition of many letters and numbers is not appreciably sacrificed if the neural network is instructed to identify those characteristics of each letter or number which are in the category "invariant." Then, rather than requiring the neural network to recognize all gradations of shape, location, size, etc. of the identified invariant characteristic, a generalized and bounded description of the invariant segments is used which requires far less inputting of sample data and less processing of information relating to an unknown letter or number.
A neural network alphanumeric pattern recognition machine comprising: means for receiving indicia of known and unknown alphanumeric symbols; a data base for storing received indicia of the known alphanumeric symbols and classifying each of said alphanumeric symbols according to a known class of an alphanumeric symbol, thereby to assign an alphanumeric class to each known alphanumeric symbol; an alphanumeric pattern classifier connected to said data base for determining alphanumeric classes of unknown symbols represented by input signals to said machine; means for training said classifier to perform a recognition function in a manner that is invariant with respect to small transformations of each said known alphanumeric symbol, said means for training comprising: means for expressing each said known alphanumeric symbol as a vector quantity; means for computing one or more tangent vectors from each said vector quantity, each said tangent vector being an indicia of a particular small transformation of said known symbol; means for combining said one or more tangent vectors, thereby to derive a combined value representative of an additional alphanumeric symbol which by definition is a member of the alphanumeric class of the known alphanumeric symbol from which derived; and means for augmenting said data base with said additional alphanumeric symbols.
Claim 2
Dependent Claim
204 words
A neural network alphanumeric pattern recognition machine comprising: means for receiving indicia of known and unknown alphanumeric symbols; a data base for storing received indicia of the known alphanumeric symbols and classifying each of said alphanumeric symbols according to a known class of an alphanumeric symbol, thereby to assign an alphanumeric class to each known alphanumeric symbol; an alphanumeric pattern classifier connected to said data base for determining alphanumeric classes of unknown symbols represented by input signals to said machine; means for training said classifier to perform a recognition function in a manner that is invariant with respect to small transformations of each said known alphanumeric symbol, said means for training comprising: means for expressing each said known alphanumeric symbol as a vector quantity; means for computing one or more tangent vectors from each said vector quantity, each said tangent vector being an indicia of a particular small transformation of said known symbol; means for combining said one or more tangent vectors, thereby to derive a combined value representative of an additional alphanumeric symbol which by definition is a member of the alphanumeric class of the known alphanumeric symbol from which derived; and means for augmenting said data base with said additional alphanumeric symbols.