15
Word Recognition Word Recognition Using Fuzzy Logic Using Fuzzy Logic 作作作作R. Buse, Z. Liu, and J. B R. Buse, Z. Liu, and J. B ezdek ezdek 作作作 作作作 作作作 作作作

Word Recognition Using Fuzzy Logic 作者: R. Buse, Z. Liu, and J. Bezdek 報告人:余家豪

Embed Size (px)

Citation preview

Word Recognition Word Recognition Using Fuzzy LogicUsing Fuzzy LogicWord Recognition Word Recognition Using Fuzzy LogicUsing Fuzzy Logic

作者:作者: R. Buse, Z. Liu, and J. BezdekR. Buse, Z. Liu, and J. Bezdek報告人:余家豪報告人:余家豪

The offline recognition of handwritten cursive word

• Segment the word into its character parts.

• Word-based.

Word-based approach’s challenges

• The complexity more greater.• Have lower discrimination

capabilities.• It is usually restricted to few word

groups.

How to extract word feature ?

• First step: Slant and Tilt Correction• Second step: Using Gabor filter• Third step: Word Alignment

Two-dimensional (2-D) fuzzy membership

function

Use to represent both sizes and

positions of the extracted word feature.

First step: Slant and Tilt

CorrectionSilver

Vegas

Second step: Using Gabor filter

The extracted feature image

Angle of Gabor filter Ø=90°

Third step: Word Alignment(1)

K-means clustering algorithm

( Centroids )

The horizontal andvertical Alignment points

Formula

Third step: Word Alignment(2)

Using alignment points to transform the extracted feature images into a standard data structure of aligned image features.

How to form fuzzy membership function

?• First step:

Composite aligned images• Second step:

Determine threshold points• Third step:

Corner point Correspondences• Fourth step:

Membership Value

First step: Composite aligned

imagesBaton

(a)

(f)(e)

(d)

(c)(b)

(g) - Rij

Second step: Determine threshold

points

Cu = 0.25

Cl = 0.1

Huij : upper threshold point

Hlij : lower threshold point

Rij

Found boundary

Third step: Corner point

Correspondences

C: set of valid combination pairs (i,j) (24 possible combinations)

T: top rectangle point

B: bottom rectangle point

|| · || distance between the two corners

The sideof theMembershipfunction

Fourth step: Membership Value

2-D membershipfunction

μ(x,y): memebership value at point (x,y)

Matching test words to the membership

function

( at Ø=90° )