Apparatus and method for dynamic font switching

Claim 1

An apparatus for recognizing a presented pattern as being in one of a plurality of predetermined classes of patterns, having generating means for generating a predetermined combination of signals representing the geometric configuration of the presented pattern, said predetermined combination of signals defining a feature vector of the presented pattern, projecting means for projecting said feature vector onto one of a plurality of sets of predetermined multi-dimensional subspaces, each of said sets characterizing a specific predetermined class of patterns, selector means for selecting one of said patterns as the presented character according to a predetermined algorithm employing the results of said projections prior to said entire projection process being completed wherein the improvement comprises: switching means operative to sense said presented pattern as being a switch character within one of said predetermined classes of patterns and in response thereto projecting said feature vector onto the set of predetermined multi-dimensional subspaces containing said switch character.

Claim 2

An apparatus for recognizing a presented pattern as being in one of a plurality of predetermined classes of patterns, having generating means for generating a predetermined combination of signals representing the geometric configuration of the presented pattern, said predetermined combination of signals defining a feature vector of the presented pattern, projecting means for projecting said feature vector onto one of a plurality of sets of predetermined multi-dimensional subspaces, each of said sets characterizing a specific predetermined class of patterns, selector means for selecting one of said patterns as the presented character according to a predetermined algorithm employing the results of said projections prior to said entire projection process being completed wherein the improvement comprises: switching means operative to sense said presented pattern as being a switch character within one of said predetermined classes of patterns and in response thereto projecting said feature vector onto the set of predetermined multi-dimensional subspaces containing said switch character.

Claim 5

An apparatus for recognizing a presented character as being in one of a plurality of predetermined classes of characters, having generating means for generating a predetermined combination of signals representing the geometric configuration of the presented character, said predetermined combination of signals defining a normalized feature vector of the presented character, storage means for storing a plurality of sets of eigenvectors, each set of eigenvectors being formed from a plurality of prototype vectors representing each class of characters, projecting means for projecting said feature vector onto a set of predetermined multi-dimensional subspaces characterizing the predetermined classes one dimension at a time, the subspaces being in the form of one of said stored sets of eigenvectors wherein each of the subspaces approximates its corresponding prototype vectors in a least mean square root sense, selector means for selecting one of said characters as the presented character according to a predetermined algorithm employing the results of said projections after each projection onto a dimension of the subspace, wherein the improvement comprises: switching means operative to sense said presented pattern as being a switch character within one of said predetermined classes of characters, and in response thereto projecting said feature vector onto the set of predetermined multi-dimensional subspaces containing said switch character.

Claim 6

An apparatus for recognizing a presented character as being in one of a plurality of predetermined classes of characters, having generating means for generating a predetermined combination of signals representing the geometric configuration of the presented character, said predetermined combination of signals defining a normalized feature vector of the presented character, storage means for storing a plurality of sets of eigenvectors, each set of eigenvectors being formed from a plurality of prototype vectors representing each class of characters, projecting means for projecting said feature vector onto a set of predetermined multi-dimensional subspaces characterizing the predetermined classes one dimension at a time, the subspaces being in the form of one of said stored sets of eigenvectors wherein each of the subspaces approximates its corresponding prototype vectors in a least mean square root sense, selector means for selecting one of said characters as the presented character according to a predetermined algorithm employing the results of said projections after each projection onto a dimension of the subspace, wherein the improvement comprises: switching means operative to sense said presented pattern as being a switch character within one of said predetermined classes of characters, and in response thereto projecting said feature vector onto the set of predetermined multi-dimensional subspaces containing said switch character.

Claim 9

A method for recognizing presented patterns each of which is one of a plurality of predetermined classes of patterns, said method comprising the steps of: generating a predetermined combination of signals each combination representing the geometric configuration of each presented pattern, the present combinations signals defining feature vectors; projecting the feature vectors onto one of a plurality of sets of predetermined multi-dimensional subspaces each set characterizing a particular predetermined class; selecting one of the patterns as the presented pattern according to a predetermined algorithm employing the results of the projections prior to the entire projection process being completed; sensing the existence of a switch character and in response thereto selecting an alternate set of predetermined multi-dimensional subspaces onto which the succeeding feature vectors are projected.

Claim 10

A method for recognizing presented patterns each of which is one of a plurality of predetermined classes of patterns, said method comprising the steps of: generating a predetermined combination of signals each combination representing the geometric configuration of each presented pattern, the present combinations signals defining feature vectors; projecting the feature vectors onto one of a plurality of sets of predetermined multi-dimensional subspaces each set characterizing a particular predetermined class; selecting one of the patterns as the presented pattern according to a predetermined algorithm employing the results of the projections prior to the entire projection process being completed; sensing the existence of a switch character and in response thereto selecting an alternate set of predetermined multi-dimensional subspaces onto which the succeeding feature vectors are projected.

Claim 11

A method for recognizing presented characters each of which is one of a plurality of predetermined classes of characters, said method comprising the steps of: generating a predetermined combination of signals, each combination representing the geometric configuration of a presented character, the generated combination of signals defining feature vectors; storing a plurality of sets of eigenvectors, each set being formed from a plurality of prototype vectors representing each class of characters; projecting each of the feature vectors onto a set of predetermined multi-dimensional subspaces characterizing the predetermined classes, one dimension at a time, such subspaces being in the form of one of said sets of eigenvectors wherein each of the subspaces approximates its corresponding prototype vectors in a least mean square error sense; selecting one of the characters as the presented character according to a predetermined algorithm employing the results of the projections prior to the entire projection process being completed; sensing the existance of a switch character, and in response thereto selecting an alternate set of predetermined multi-dimensional subspaces onto which the succeeding feature vectors are projected.

Claim 12

A method for recognizing presented characters each of which is one of a plurality of predetermined classes of characters, said method comprising the steps of: generating a predetermined combination of signals, each combination representing the geometric configuration of a presented character, the generated combination of signals defining feature vectors; storing a plurality of sets of eigenvectors, each set being formed from a plurality of prototype vectors representing each class of characters; projecting each of the feature vectors onto a set of predetermined multi-dimensional subspaces characterizing the predetermined classes, one dimension at a time, such subspaces being in the form of one of said sets of eigenvectors wherein each of the subspaces approximates its corresponding prototype vectors in a least mean square error sense; selecting one of the characters as the presented character according to a predetermined algorithm employing the results of the projections prior to the entire projection process being completed; sensing the existance of a switch character, and in response thereto selecting an alternate set of predetermined multi-dimensional subspaces onto which the succeeding feature vectors are projected.