Image analysis – Pattern recognition – On-line recognition of handwritten characters
Reexamination Certificate
1996-10-11
2001-03-27
Couso, Yon J. (Department: 2623)
Image analysis
Pattern recognition
On-line recognition of handwritten characters
C382S201000
Reexamination Certificate
active
06208757
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to the processing of human handwriting for purposes of information storage, automatic character recognition, and the like. More particularly, this invention relates to the processing of spatiotemporally sampled symbols, including the representation of such symbols in parametric form and the reconstruction of such symbols from their parameters to create legible characters.
ART BACKGROUND
The automatic analysis of human handwriting begins with sampling and digitization of the image or signal produced directly by human manipulation of a writing instrument, referred to herein, in a general sense, as a stylus. Purely graphical analysis can be performed on data sampled from a static image. However, for dynamic analysis, and more generally for analyzing data with reference to a temporal sequence, it is advantageous for the human subject to produce data by manipulating an instrumented stylus or tablet that permits spatiotemporal sampling.
One exemplary instrumented tablet is described in U.S. Pat. No. 5,463,388, issued on Oct. 31, 1995 to R. A. Boie et al. This tablet includes a rectangular array of capacitance-sensing electrodes. The position of a handheld stylus is determined, e.g., from the centroid of the respective capacitance values, as calculated in a microcontroller.
Parametric methods have been used for a number of years in connection, for example, with automatic signature verification. According to these methods, the signature (or other handwritten symbol) is represented in an abstract parameter space. The parametric representation consists of a set of numerical values of functions that are evaluated on the sampled data, and that relate to some combination of graphical and dynamic properties of the sampled data. Generally, a parametric representation is a condensed representation, in the sense that it occupies fewer bits of data-storage capacity than do the raw, sampled data.
Parametric representations of signatures have been used with some success for signature verification. In signature verification, the parameters are evaluated on a newly entered signature (or group of signatures), and the results are compared with a stored set of reference values. Such a procedure does not require the reconstruction, from parameters, of either the reference signature or the newly entered signature. Therefore, there is no need to choose parameters that preserve enough graphical information to reconstruct these signatures. Instead, parameters for signature verification are selected on the basis, e.g., of a tradeoff between selectivity and computational efficiency.
However, there is a need for a compact method for storing handwritten symbols that can later be reconstructed as legible characters. For such a purpose, it is not sufficient to use the abstract parameters that are often used in the context of signature verification. On the other hand, the more detailed the shape information that is used, the less compact the resulting storage method will be.
What the field has lacked, until now, are methods for storing and reconstructing characters that enjoy the benefits of compact parametric techniques, but still lead to clearly legible reconstructed characters.
SUMMARY OF THE INVENTION
I have invented such a reconstruction method, based on appropriate parametric methods of data storage.
An appropriate way of segmenting handwriting into “strokes” and then representing each such stroke by a set of values for a small number of parameters (the “standard parameter set”) is described in my co-pending U.S. patent application Ser. No. 08/729,463 “Method And Apparatus For Parametric Representation Of Handwritten Symbols,” which is hereby incorporated by reference. The strokes generated by the method therein described are free of major inflection points (i.e., points at which the curvature changes sign). Taking account of this property, the stroke reconstruction method described below always produces reconstructed strokes that are entirely free of inflection points. Because minor inflection points generated by noise and waviness are lost, the reconstructed strokes are often smoother than the originals. Although those skilled in the art could modify the method described below to accommodate major inflection points, I have found it preferable on grounds of simplicity to segment the handwriting so they do not occur inside of strokes.
The standard parameter set for a particular stroke comprises the x and y coordinates of the stroke's initial and final endpoints, the arc length s of the stroke, the initial tangent angle &thgr;, and the net turning angle &phgr;.
The initial tangent angle &thgr; is defined with reference to the line through the endpoints; it is the angle from this line to the stroke's tangent at the initial endpoint.
The net turning angle &phgr; is defined with reference to a moving tangent. If the tangent is evisaged as a rigid rod lying against the stroke, then &phgr; is the net angle through which the rod turns as the point of tangency traverse the stroke from its initial to its final endpoint. (In accordance with mathematical convention, counterclockwise turning is counted as positive and clockwise turning is counted as negative.) Since the stroke is free of major inflection points, the tangent turns primarily in one direction (clockwise or counterclockwise) during its traversal of the stroke.
The present invention involves reconstruction of legible symbols from parametric data of the kind described above. These data are preferably derived using the methods described in U.S. patent application Ser. No. 08/729,463 cited above, but the use of other methods to derive the same or similar data will not diminish the usefulness of the present invention for reconstructing legible symbols therefrom.
Briefly, the present invention in a broad aspect involves reconstructing one or more strokes from their respective standard parameter sets by drawing each stroke as a sequence of smoothly joined straight lines and circular arcs of common radius. The largest number of segments (lines and arcs) needed to reconstruct any stroke is three. The segments are found by a simple algorithm.
REFERENCES:
patent: 4550438 (1985-10-01), Convis et al.
patent: 5185818 (1993-02-01), Warnock
patent: 5347589 (1994-09-01), Meeks et al.
patent: 5559897 (1996-09-01), Brown et al.
patent: 5577135 (1996-11-01), Grajski et al.
patent: 5596656 (1997-01-01), Goldberg
patent: 5698822 (1997-12-01), Haneda et al.
patent: 5761328 (1998-06-01), Solberg et al.
Couso Yon J.
Lucent Technologies - Inc.
LandOfFree
Method and apparatus for reconstructing handwritten symbols... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for reconstructing handwritten symbols..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for reconstructing handwritten symbols... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2464167