Presentation is loading. Please wait.

Presentation is loading. Please wait.

Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 1 Notes on Contemporary Table Recognition David W. Embley 1,

Similar presentations


Presentation on theme: "Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 1 Notes on Contemporary Table Recognition David W. Embley 1,"— Presentation transcript:

1 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 1 Notes on Contemporary Table Recognition David W. Embley 1, Daniel Lopresti 2, and George Nagy 3 2 Department of Computer Science and Engineering, Lehigh University, Bethlehem, PA Department of Electrical, Computer, and Systems Engineering, Rensselaer Polytechnic Institute, Troy, NY Department of Computer Science, Brigham Young University, Provo, UT 84602

2 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 2 Table I falder multon burlam hepth (gd) gonsity (ld/gg)fleck

3 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 3 Table Layout Analysis xxx xxxxx xxx xxxxx xxx xxxxx xxxxx xxx xxxxxxx xxxx xxxxx

4 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 4 Array Models for Table Analysis

5 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 5 Table Transcription falder multon burlam hepth (gd) gonsity (ld/gg) fleck

6 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 6 Table II 230 gd2.3 ld/ggfalder 350 gd2.5 ld/ggelmer 120 gd1.2 ld/gggoldam Table transcription is not sufficient for combining information from Table I and Table II. But it can often be accomplished with current commercial software.

7 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 7 Existing Table Conversion Software HTML web pages MS-Word table creation Adobe PDF web pages Array model geometric representation XML tags TIF scanned bitmap Wang model abstract table * Access relational DBMS MS-Excel spreadsheet * Tabular abstraction, editing, and formatting, Xinxin Wang, PhD thesis, University of Waterloo, 1996.

8 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 8 Existing Table Conversion Software Table as rendered by Microsoft Internet Explorer 6.0:

9 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 9 Existing Table Conversion Software Table copied into Microsoft Word 10.2:

10 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 10 Existing Table Conversion Software Table copied from Microsoft Word into Excel 10.2:

11 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 11 Existing Table Conversion Software ASCII version of the table (as rendered by MS Word):

12 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 12 Existing Table Conversion Software Table rendered from a PDF file:

13 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 13 Table Interpretation fleck virtual column header (“characteristic”) falder multon burlam hepth (gd) gonsity (ld/gg) (stub)

14 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 14 “Layout-invariant” Wang Notation Essentially a set of category trees with common leaf cells – requires modification for table interpretation. Categories  =({fleck.burlam, characteristic.gonsity})→ 1.2 ({fleck.falder, characteristic.gonsity}) → 2.3 … ({fleck.multon, characteristic.hepth}) → 350 Header cell mappings C =(fleck,{(bulram,  ), (falder,  ),(multon,  )}), (characteristic, {(gonsity,  ), (hepth,  )}).

15 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 15 Table III hepth (ld/cg) gonsity (ld/cg)virtual row header multonfalderburlam fleck Same Wang Notation as Table I.

16 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 16 On Human Interaction  Complete automation is still out of reach.  Some HCI is unavoidable, sooner or later.  Sooner is better than later.  Discovering errors later increases negative consequences (and potential for embarrassment).  Partial automation + interaction is more accurate than machine alone and faster than unaided human.  TabbyCat (for “table categorizer”) is our prototype interactive tool for supporting Wang-style markup.

17 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 17 Table Understanding  Currently multon is the best value for rapitting velters. It is about 25% better than burlam or falder, which are nearly the same.  Check Table II to see whether elmer is even better. With domain knowledge? Ontology??

18 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 18 Table Annotation & Truthing Snapshot of tool for creating Wang-style table mark-up:

19 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 19 Table Annotation & Truthing Another snapshot of Wang-style table mark-up tool:

20 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 20 Current Status  Most previous work (>100 papers) on table layout analysis and transcription.  Easier for tables in symbolic form (.doc,.xls,.pdf, html) than for scanned bitmaps.  Most methods process only one table at a time.  Commercial software already does it fairly well.  Interaction required in many cases.  The next step, in our view, is table interpretation.  Table understanding is waiting in the wings.

21 Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 21 Thank you! Photograph taken near Lehigh University yesterday (no snow when we left for NZ):


Download ppt "Notes on Contemporary Table Recognition Embley, Lopresti, and Nagy  February 2006  Slide 1 Notes on Contemporary Table Recognition David W. Embley 1,"

Similar presentations


Ads by Google