Handwriting recognition system and method

A handwritten character recognizer system for recognizing individual characters from a character set comprising:

• means for detecting character strokes in real-time as they are being written;
• means for segmenting said character strokes into individual characters;
• means for measuring stroke geometry at a plurality of points of ones of said character strokes comprising individual characters;
• means for producing sparse fuzzy membership vectors representing said stroke geometry of a plurality of points of ones of said character strokes comprising individual characters;
• means for combining said sparse fuzzy membership vectors into features;
• means for collecting said features into a sparse data structure;
• and
• a statistical classifier system having inputs for receiving said sparse data structure and outputs, said system being capable of determining the character before all strokes in the character are detected.

A system for recognizing handwritten characters having a plurality of strokes detected in real-time as they are being written, said system comprising a statistical classifier, said statistical classifier comprising:

• a set of chunks, each said chunk comprising a set of weights;
• an input port coupled to a sparse data structure, said sparse data structure comprising a set of substructures corresponding to said set of chunks;
• means for applying said weights in one said chunk to a corresponding substructure in said sparse data structure and to generate a set of terms;
• and
• an accumulator coupled to said terms, said system being capable of determining the character before all strokes in the character are detected.

A system for recognizing handwritten characters detected in real-time as they are being written, said system comprising:

• first apparatus disposed to receive said handwritten character;
• a preprocessor coupled to said first apparatus and disposed to generate a set of fuzzy membership values;
• second apparatus coupled to said fuzzy membership values and disposed to generate a set of feature pairs of data identifiers and data values in response thereto;
• and
• a statistical classifier coupled to said feature pairs and disposed to apply a set of weights only to said data values corresponding to said data identifiers, said system being capable of determining the character before all strokes in the character are detected.

Apparatus for supplying data to a handwritten recognizer system comprising:

• a detector for detecting character strokes in real-time as they are being made;
• a segmenting device for segmenting said character strokes into individual characters;
• a measuring device for measuring stroke geometry at a plurality of points of ones of said character strokes comprising individual characters;
• a device for producing sparse fuzzy membership vectors into features;
• a combining device for combining said sparse fuzzy membership vectors into features;
• a collector for collecting said features into a sparse data structure;
• and
• a supplier for supplying said sparse data structure to a handwritten recognition system, said apparatus being capable of determining the character before all strokes in the character are detected.

A handwritten character recognizer system for recognizing individual characters from a character set comprising:

• a detector for detecting character strokes in real-time as they are being made;
• a segmenting device for segmenting said character strokes into individual characters;
• a measurement device for measuring stroke geometry at a plurality of points of ones of said character strokes comprising individual characters;
• a producing device for producing sparse fuzzy membership vectors representing said stroke geometry of a plurality of points of ones of said character strokes comprising individual characters;
• a combiner for combining said sparse fuzzy membership vectors into features;
• a collector for collecting said features into a sparse data structure;
• and
• a statistical classifier system having inputs for receiving said sparse data structure and outputs, said system being capable of determining the character before all strokes in the character are detected.

A system for recognizing handwritten characters having a plurality of strokes detected in real-time as they are being made, said system comprising a statistical classifier, said statistical classifier comprising:

• a set of chunks, each said chunk comprising a set of weights;
• an input port coupled to a sparse data structure, said sparse data structure comprising a set of substructures corresponding to said set of chunks;
• an applying device for applying said weights in one said chunk to a corresponding substructure in said sparse data structure and to generate a set of terms;
• and
• an accumulator coupled to said terms, said system being capable of determining the character before all strokes in the character are detected.