41
Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Embed Size (px)

DESCRIPTION

Why do vectorization? A lot of old drawings to be reused, and CAD files are more editable than images. Preprocess of automatic drawing understanding systems (information statistic, 3D reconstruction). Save the storage space.

Citation preview

Page 1: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Introduction to Vectorization of Engineering Drawings

Song Jiqiang18/9/2001

Page 2: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Concept

• Vectorization: the conversion from a raster image to its vector-form file.

Engineering drawings

in paper form

Raster images

Vector-form(CAD files)

Scan Vectorization

Page 3: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Why do vectorization?

• A lot of old drawings to be reused, and CAD files are more editable than images.

• Preprocess of automatic drawing understanding systems (information statistic, 3D reconstruction).

• Save the storage space.

Page 4: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

History• Begin research on vectorization at late 70’s;

• Begin that for engineering drawings at late

80’s;• Related organizations & publications

– IAPR TC-10, IEEE– IEEE T.PAMI, PR, PRL, CVIU, CVGIP, etc.– ICPR, ICDAR, IEEE CVPR, etc.

Page 5: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

State of arts

• K. Tombre (LNCS vol.1389, 1998) : “None of these methods works. … Actually, the m

ethods do work, but none of them is perfect.”

• 《 CADALYST 》 performs the annual evaluation on commercial vectorization systems.

Page 6: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Review of existing methods

• Thinning based • CT(Contour Tracking) based• RLE(Run Length Encoding) based• SPT(Sparse Pixel Tracking) based

Page 7: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Thinning-based methods

Page 8: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

CT-based methods

Page 9: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

RLE-based methods

Page 10: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

SPT-based methods

P3

P0

P1

P2P0 P1 P2 P3 P4

Page 11: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Existing difficulties

• Lines with intersections broken into pieces.

• Texts touch lines misrecognition.

• The interference of recognized objects repetitive detection, false detection.

Page 12: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Our research

• Analysis the vectorization model of existing methods.

• Propose an efficient vectorization model for engineering drawings.

• Propose a group of new graphical object recognition algorithms.

Page 13: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Common model of existing methods ( 2PV - 2 Phase Vectorization )

Raster image

Low level vector form:Short line segments

Graphical objects:Straight lines, circle/arc, curve

Phase 1: Skeletonization or medial-axis approach

Phase 2: Vector-based graphical object recognition

Page 14: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Motivation of 2PV

• Internal memory (RAM)– Used to be high price & limit capacity– High pixel access frequency cause swap

• Pixel tracking algorithm– No guide direction– Repetitive tracking

Page 15: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Object-Oriented Progressive-Simplification based Vectorization Model

• 1 phase model– Imitate the way that humans read drawings

• Recognize a graphical object in its entirety– Object-oriented feature

• Simplify the image data as the recognition goes on– Progressive-simplification feature

Page 16: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Workflow of OOPSV

S-LineRecognition

ArcRecognition

CurveRecognition

SymbolRecognition

TextRecognition

LineSymbolText

ArcCurveSymbolText

CurveSymbolText

SymbolText

Text

S-Line Arc Curve Symbol Text

Page 17: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Graphical object recognition

• Get the intrinsic characteristic of individual type of graphical object.

• Use the characteristic as a guide to track the graphic object in complex environment.

Page 18: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Straight line recognition

Seed segment detection

A

Seed Segment

Irregular run Regular run

Page 19: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Direction guided tracking — based on the Bresenham algorithm

Straight line recognition

O

Vo Vp

Seed segment

Perpendicular runs

Black segment

White segment

Page 20: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Straight line recognition

Dynamic adjustment to tracking direction

PP’

Vp

O

P

P’

Vp

O Pe

Page 21: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Line net recognition

• A line net is a group of intersecting lines.• Take advantage of the intersecting

relationship to accelerate recognition.• Example:

Page 22: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Circle/Arc recognition• Arc segment detection

– get initial arc center, radius, thickness

• Circular tracking– based on the Bresenham algorithm for circle

Page 23: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Circular tracking

P2

P1

O

P

P’

PE

(a) Adjust succeeds (b) Adjust fails

Legend: raster image tracking path medial axis testing path

P2

P1

O P P’

PE

Page 24: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Curve tracking

Tracking result: a sequence of polyline.

P5

P4

P1P2 P3

P8

P9

P7

P6

Pm

P10P11

Page 25: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Image simplification

• Intersection-preserving pixel deletion• Based on the contour detection of the

intersecting branches

Branch at one side

MiddleLine

Branches at both sides

RecognizedLine

Contours of branches

Page 26: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Symbol recognition

• Common symbols– Cartography-based recognition

• Domain-specific symbols– Template-based recognition

Page 27: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Cartography

Page 28: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Symbol template

Page 29: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Text recognition

• Text segmentation– Difficulties: text touches line, similar size

• Character recognition– Stroke-based recognition algorithm

Page 30: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Image before the line recognition

Page 31: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Image after the line recognition

Page 32: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Suspension-Release mechanism

Condition 1: Size( Box(li) ) < Tl Condition 2: p, pl C(p,L) >> C( FP(l,L), L)

n

i=1

Page 33: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Stroke-based character template

• Stroke definition• Black position• White position• Aspect ratio scope• Complexity level

Page 34: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Character recognition

(UMNKLDREFBbhklI1) (BE)

(E) Accept ‘E’, then cut image

Page 35: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Separate connected characters

a. When the rightmost stroke is vertical

b. When the rightmost stroke is not vertical

f(x) x1

•Base on the analysis of the rightmost stroke and the vertical projection.

Page 36: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Experimental result

• Implemented a complete vectorization system running on Windows platform using VC6.0.

• Automatic vectorization of an A0-size drawing (15M) takes about 5 minutes. (PIII500/128M)– Line vectorization takes less than 1 minute (1600 lines),

faster than performing a thinning operation (3.5 mins).

Page 37: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

CDI evaluationSize Dp Fp PRI Dv Fv VRI CDI

A0 0.966 0.086 0.940 0.954 0.082 0.936 0.938

A1 0.973 0.077 0.948 0.965 0.084 0.941 0.945

A2 0.975 0.054 0.961 0.963 0.068 0.948 0.955

A3 0.962 0.049 0.957 0.959 0.066 0.947 0.952

This protocol was proposed in Machine Vision Application, (1997)

Page 38: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Manual editing cost evaluationDrawingSize

False_alarms Misses One2many Many2one Total Costs

A0 Ⅰ 124 80 24 37 354Ⅱ 92 62 16 32 263

A1 Ⅰ 63 46 11 18 194Ⅱ 41 33 9 16 142

A2 Ⅰ 30 26 6 8 101Ⅱ 22 18 3 6 67

A3 Ⅰ 16 15 5 7 69Ⅱ 10 13 3 4 44

This protocol was proposed in LNCS V.1389, (1998)

Page 39: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Comparison of editing cost with VPStudio

a. raster image b. our system c. VPStudio

Conclusion: Object-oriented recognition algorithms produce less misrecognition, therefore the editing cost has decreased.

Page 40: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Conclusion

• Progressive simplification decreases both the complexity and workload of vectorization.

• The object-oriented recognition algorithms recognize graphical objects fast and entirely.

Page 41: Introduction to Vectorization of Engineering Drawings Song Jiqiang 18/9/2001

Related papers• 7 journal papers & 4 conference papers

• IEEE Trans. PAMI reviewer’s comment: “An efficient model is very important to

recognize engineering drawings …. This paper suggested an object-oriented

progressive-simplification based vectorization system for engineering drawings.

It would bring an impact in this area.”