EP0562767A2
Expired / Lapsed
EP
⏱️ 0.0 years remaining
Method of recognizing handwritten symbols
Assignee: AT&T Inc
•
Filing: 1993-03-18
•
Granted: —
•
Expires: 2013-03-18
Abstract
Abstract An unknown handwritten symbol written on a digitizing tablet is compared with symbols in a predefined `alphabet' or library of model symbols and the closest match chosen. Recognition is independent of the size, position or orientation of the symbols. The alphabet can be any collection of symbols, such as alphanumeric characters, ideograms or words in cursive script and is created by writing at least one example of each symbol on the tablet. A sequence of samples of the pen position is recorded while a symbol is being written. The samples form a vector, which is then translated (45) so that the centroid of the symbol lies at an origin. The comparison, which can easily be done in real time, involves calculating (46) a correlation factor from scalar products of the vector (v α ) for the unknown symbol and two versions (v k and u k ) of the vector for each model symbol and choosing (49) the model symbol having the highest correlation factor. If needed to distinguish between model symbols with similar correlation factors, the comparison can also include calculating a rotation factor from such vectors for use in making such choice. Embodiments of the invention can be configured that are user-independent, user-dependent or that evolve from one to the other.
Inventors
- 👤 Frank William Sinden
- 👤 Gordon Thomas Wilfong
Bibliographic Details
Application Number
—
Jurisdiction
EP
Data Source
google_patents
Last Synchronized
2026-11-09 12:15
CPC Classifications
G; G07; G07C; G07C9/00; G07C9/30; G07C9/32; G07C9/35; G06; G06V; G06V30/00; G06V30/10; G06V30/32; G06V30/333; G06V30/36; G06V30/373
Claims Analysis
Total Claims: 17
Claim 1
Independent Claim
162 words
A method of finding the closest match for an unknown symbol in a set of stored model symbols wherein said unknown symbol and said model symbols are each represented as an ordered set of coordinate samples with the centroids of said unknown symbol and said model symbol at a common origin and wherein said samples for said unknown symbol are compared in turn with said samples for each of said model symbols, CHARATERIZED IN THAT said method further comprises: before each comparison, equalizing the number of translated samples for said unknown symbol and said model symbol during each said comparison, calculating the correlation of the equalized number of said translated samples for said unknown symbol with the equalized number of said translated samples for said model symbol and identifying the model symbol associated with the highest correlation as the closest match, whereby said comparisons are independent of the size, position or orientation of said unknown symbol with respect to said model symbols.
Claim 2
Dependent Claim
83 words
The method of claim 1 FURTHER CHARACTERIZED IN THAT said calculating step further comprises calculating said correlation as: where vector v α is the equalized ordered set of translated samples for said unknown symbol, vector v k is the equalized ordered set of translated samples for said model symbol, vector u k is derived from vector v k by a counterclockwise rotation of π/2 radians and said symbol associated with the highest correlation is the symbol associated with the highest value of C.
Claim 3
Dependent Claim
35 words
The method of claim 1 FURTHER CHARACTERIZED IN THAT said method further comprises: storing said ordered sets of coordinate samples for said model symbols while said model symbols are being handwritten on a digitizing tablet.
Claim 4
Dependent Claim
57 words
The method of claim 3 FURTHER CHARACTERIZED IN THAT said method further comprises: storing the number of pen strokes used to write each model symbol and each unknown symbol and wherein said comparing step further comprises: limiting said comparing step to those of said model symbols having the same number of pen strokes as the unknown symbol.
Claim 5
Dependent Claim
21 words
The method of claim 4 FURTHER CHARACTERIZED IN THAT said equalizing step further comprises equalizing said samples on a stroke-by-stroke basis.
Claim 6
Dependent Claim
66 words
The method of claim 5 FURTHER CHARACTERIZED IN THAT said equalizing step further comprises: limiting vector v α to a subset of essentially uniformly spaced samples from said ordered set for said unknown symbol, and limiting vector v k to include the same number of essentially uniformly spaced samples, stroke-by-stroke, from said ordered set for said model symbol as in said limited vector v α .
Claim 7
Dependent Claim
57 words
The method of claim 1 FURTHER CHARACTERIZED IN THAT if an unknown symbol is recognized incorrectly, adapting said set of model symbols by adding said ordered sequence of samples for said incorrectly recognized symbol to said set of model symbols as an additional representative example of the correct symbol to be associated with said incorrectly recognized symbol.
Claim 8
Dependent Claim
43 words
The method of claim 1 FURTHER CHARACTERIZED IN THAT said method further comprises continually adapting said set of model symbols by storing said ordered sequences of samples for each of a specified number of the most recently entered instances of each model symbol.
Claim 9
Dependent Claim
84 words
The method of claim 2 FURTHER CHARACTERIZED IN THAT said method further comprises: establishing a threshold value related to said highest value of C, if the value of C for more than one model symbol exceed said threshold value then, for each model symbol having a value of C above said threshold value, calculating the rotation of said model symbol with respect to said unknown symbol and wherein said recognizing step comprises recognizing said unknown symbol as the model symbol having the smallest rotation.
Claim 10
Dependent Claim
55 words
The model of claim 9 FURTHER CHARACTERIZED IN THAT said rotation calculating step further comprises: calculating said rotation as: R = v α · v k (v α · v k )² + (v α · u k )² wherein said model symbol having the smallest rotation is the symbol having R closest to unity.
Claim 11
Dependent Claim
28 words
The method of claim 10 FURTHER CHARACTERIZED IN THAT said establishing step further comprises establishing said treshold value as a fixed percentage of said highest value of C.
Claim 12
Dependent Claim
34 words
The method of claim 3 FURTHER CHARACTERIZED IN THAT said method is used for verifying signatures, said stored model symbols are previously written signatures and said unknown symbol is a signature to be verified.
Claim 13
Dependent Claim
44 words
The method of claim 12 FURTHER CHARACTERIZED IN THAT said method further comprises: comparing said highest value of C with the next-highest value of C and rejecting said signature to be verified if said next-highest value exceeds a certain percentage of said highest value.
Claim 14
Dependent Claim
36 words
The method of claim 1 FURTHER CHARACTERIZED IN THAT said method further comprises: storing sets of pre-recorded coordinate samples for a plurality of versions of each said model symbol in said set of stored model symbols.
Claim 15
Dependent Claim
47 words
The method of claim 14 FURTHER CHARACTERIZED IN THAT said method further comprises: as unknown symbols are recognized, replacing at least part of said pre-recorded samples for each said model symbol with at least one set of samples for an unknown symbol recognized as said model symbol.
Claim 16
Dependent Claim
41 words
The method of claim 15 FURTHER CHARACTERIZED IN THAT said replacing step further comprises: replacing said sets of pre-recorded samples for each said model symbol with a smaller number of sets of samples for unknown symbols recognized as said model symbol.
Claim 17
Dependent Claim
45 words
The method of claim 14 FURTHER CHARACTERIZED IN THAT said method further comprises as unknown symbols are recognized augmenting said sets of prerecorded samples for each said model symbol with at least one set of samples for an unknown symbol recognized as said model symbol.
References Cited (Backward Citations)
No citation records indexed.
USPTO Annuity & Maintenance Windows
Maintenance schedule is calculated for granted US utility patents.