Table detection in ink notes
AssigneeMicrosoft Corporation
InventorsZhouchen Lin, Junfeng He, Zhicheng Zhong, Chun-Hui Hu, Jian Wang
Life of this patent
Estimating the term…
Abstract
Computer-readable media having computer-executable instructions and apparatuses detect a table in a handwritten document. Line segments are derived from drawing strokes so that a bounding frame of a candidate table is obtained. An associated table structure is consequently recognized from the bounding frame, lines segments within the bounding frame, and their intersection points. A classifier that reflects at least one table characteristic is determined, and the candidate table is consequently validated or rejected.
Claims20
A computer-readable medium storing computer-executable instructions that, when executed, cause one or more processors to perform steps comprising:
- (a) determining a line list from a handwritten document;
- (b) obtaining a bounding frame of a candidate table from the line list;
- (c) recognizing a structure of the candidate table from the bounding frame and the line list;
- and
- (d) validating the candidate table from a classifier to determine whether the candidate table corresponds to a validated table.
A table detector that detects a table in a handwritten document, the table detector comprising:
- a preprocessor that forms a line list from a line block of the handwritten document by approximating a drawing stroke with conjoined line segments;
- a feature extractor that determines a bounding frame for a candidate table, recognizes a structure of the candidate table from the bounding frame and a line list, analyzes content within the candidate table, and determines confidence levels associated with the bounding frame, the structure, and the content;
- and
- a classification module that forms a classifier from the confidence levels, whether the candidate table corresponds to a validated table when the classifier is larger than a predetermined threshold, and that outputs a semantic table for the validated table.
A computer-readable medium having computer-executable instructions for performing steps comprising:
- (a) determining a line list from a handwritten document;
- (b) obtaining a bounding frame from the line list;
- (c) forming a candidate table from the bounding frame and line segments of the line list that are interior to the bounding frame;
- (d) normalizing the candidate table by replacing approximately horizontal line segments with essentially horizontal line segments and by replacing approximately vertical line segments with essentially vertical line segments.
- (e) recognizing a structure of the candidate table from intersection types of the candidate table;
- (f) analyzing content in the candidate table from a crossing of a stroke and a line segment of the candidate table;
- (g) determining a classifier that provides a level of confidence whether the candidate table is a valid table;
- and
- (h) when the classifier is larger than a predetermined threshold, outputting a semantic table.
Dependencies are inferred from the claim text.