Handwriting recognition systems and methods

1. A method for recognizing handwriting, the method comprising:

receiving handwriting data;

determining geometric data of text in handwriting data;

determining sub-characters of the text;

matching sub-characters of text to a model; and

determining most probable characters of the text based on said matching,

wherein determining geometric data of text in handwritten data comprises:

estimating extrinsic geometry of text defined by a translation and a rotation of a local portion of a line of the received handwriting data by estimating baseline of text (including tilt) from line segments that are estimated tangent to pairs of y-infima of character strokes and tangent to collinear triplets from y-infima.

2. The method of claim 1 ,

wherein determining intrinsic geometry of the handwritten text defined by size and shape of characters and sub-characters comprises:

estimating slant angle of the text relative to the baseline;

estimating waist, ascender, and descender lines parallel to the baseline in lowercase text, or caps line in block caps text; and

estimating these lines parallel to the baseline at the local optimal robust least squares estimate of y-suprema and y-infima of the strokes that are estimates of the local averages of these lines;

wherein determining sub-characters of the text comprises:

determining sub-characters based on ovals, loops, retraces, lobes, and sticks;

separating stroke segments from sequences of blended strokes;

estimating these separate stroke segments pairwise at natural transitions based on the sub-character components; and

representing the geometric data of sub-characters as free knot splines parameterized by arc length, with spiral bases with C2 continuity at knots, and with C0 continuity at discontinuities,

wherein matching sub-characters of text to a model comprises:

mapping the model to character instances by curves, pairs of curves, and pairs of pairs of curves,

mapping structured models of characters composed of sub-characters; and

representing the mapped sub-characters and characters as a directed acyclic graph (DAG).

3. The method of claim 2 wherein the model includes co-articulation of every pair of characters to make use of their orthographic evidence to reduce their variance.

4. The method of claim 2 , further comprising estimating the probability of a pair or tuple of received characters based on sub-word linguistic evidence of graphemes, syllables and subsyllables, and affixes.

5. A method for recognizing handwriting, the method comprising:

receiving handwriting data;

determining geometric data of text in handwriting data;

determining sub-characters of the text;

matching sub-characters of text to a model; and

determining most probable characters of the text based on said matching,

wherein determining geometric data of text in handwriting data includes determining intrinsic geometry of the handwritten text defined by size and shape of characters and sub-characters,

wherein determining intrinsic geometry of the handwritten text defined by size and shape of characters and sub-characters comprises:

estimating slant angle of the text relative to the baseline based on a local estimate of skew of circles and straight lines;

estimating waist, ascender, and descender lines parallel to the baseline in lowercase text, or caps line in block caps text; and

estimating these lines parallel to the baseline at the local optimal robust optimization estimate of y-suprema and y-infima of the strokes that are estimates of the local averages of these lines.

6. A non-transitory computer-readable storage medium storing instructions for handwriting recognition, the instructions when executed by one or more processors causing the one or more processors to perform steps comprising:

receiving handwriting data;

determining geometric data of text in handwriting data;

determining sub-characters of the text;

matching sub-characters of text to a model; and

determining most probable characters of the text based on said matching,

wherein determining geometric data in text of handwritten data comprises:

estimating extrinsic geometry of text defined by a translation and a rotation of a local portion of a line of the received handwriting data by estimating baseline of text (including tilt) from line segments that are estimated tangent to pairs of y-infima of character strokes and tangent to collinear triplets from y-infima.

7. A computer system for handwriting recognition, the computer system comprising: one or more computer processors; and one or more non-transitory computer-readable storage media, the storage media storing computer program instructions executable by the one or more computer processors to perform steps comprising:

receiving handwriting data;

determining geometric data of text in handwriting data;

determining sub-characters of the text;

matching sub-characters of text to a model; and

determining most probable characters of the text based on said matching, wherein determining geometric data in text of handwritten data comprises:

estimating extrinsic geometry of text defined by a translation and a rotation of a local portion of a line of the received handwriting data by estimating baseline of text (including tilt) from line segments that are estimated tangent to pairs of y-infima of character strokes and tangent to collinear triplets from y-infima.

8. A non-transitory computer-readable storage medium storing instructions for handwriting recognition, the instructions when executed by one or more processors causing the one or more processors to perform steps comprising:

receiving handwriting data;

determining geometric data of text in handwriting data;

determining sub-characters of the text;

matching sub-characters of text to a model; and

determining most probable characters of the text based on said matching,

wherein determining geometric data of text in handwriting data includes determining intrinsic geometry of the handwritten text defined by size and shape of characters and sub-characters,

wherein determining intrinsic geometry of the handwritten text defined by size and shape of characters and sub-characters comprises:

estimating slant angle of the text relative to the baseline based on a local estimate of skew of circles and straight lines;

estimating waist, ascender, and descender lines parallel to the baseline in lowercase text, or caps line in block caps text; and

estimating these lines parallel to the baseline at the local optimal robust optimization estimate of y-suprema and y-infima of the strokes that are estimates of the local averages of these lines.

9. A computer system for handwriting recognition, the computer system comprising: one or more computer processors; and one or more non-transitory computer-readable storage media, the storage media storing computer program instructions executable by the one or more computer processors to perform steps comprising:

receiving handwriting data;

determining geometric data of text in handwriting data;

determining sub-characters of the text;

matching sub-characters of text to a model; and

determining most probable characters of the text based on said matching,

wherein determining geometric data of text in handwriting data includes determining intrinsic geometry of the handwritten text defined by size and shape of characters and sub-characters,

wherein determining intrinsic geometry of the handwritten text defined by size and shape of characters and sub-characters comprises:

estimating slant angle of the text relative to the baseline based on a local estimate of skew of circles and straight lines;

estimating waist, ascender, and descender lines parallel to the baseline in lowercase text, or caps line in block caps text; and

estimating these lines parallel to the baseline at the local optimal robust optimization estimate of y-suprema and y-infima of the strokes that are estimates of the local averages of these lines.