Transcript Progress presentation

Single Character Recognition
CS5185-MULTIMEDIA TECHNOLOGIES AND APPLICATIONS
2007 1st Semester - Group Project Progress Presentation
Presented on 2007-10-23
Project Group
Group 08
 Members:

Ku Heung Chin (Ku)
 Yu Kam Fung (Kam)
 Fung Ka Hang (Harry)
 Wang Yang

Usage of Content Recognition

Printed content
Business Card Recognition
 Car License plate recognition


Handwritten content
Postal address recognition
 Bank cheque signature verification

A top-down approach
Document Image
Word Phrases
Word Phrases
Word Phrases
Single
Character
Single
Character
Single
Character
Single Character Recognition
is the basis for
Content Recognition !!
Implementation

Programming Language

Java (J2SE 1.6)
Platform independent
 Rich graphics and image processing APIs
 Vector data structure APIs


Development platform

Eclipse - Java SDK
User friendly interface
 Java complier and debugger
 Java Syntax and spelling checking

How does it work?
Character DB
Preprocessing
A
…
G
…
Recognition
G
Preprocessing
Image to *.bmp format
 Color(RGB) image to Grayscale image
 Noise Filter
 Binarization
 Slant Correction
 Thinning
 Size Normalization

Image to *.bmp format
FileChooser Class to select input image.
 Accepted different image format, e.g. jpg,
gif, etc…
 Call some API to convert to 24-bit RGB
Bitmap format (will be implemented in
later phase)

RGB to 8-bits Grayscale
0.3*Red
+
0.59*Green
+
0.11*Blue
Noise Filter

3x3 pixels Median Filter
0
100 250
40
10
120
80
90
20
Median:80
0
100
250
40
80
120
80
90
20
Binarization
Purpose :
 Clear cut the background and the
character
Binarization
Implementation (Simplest Approach) :
 Assumption that noises are filtered out
and dark text and white background
 Transform gray color to black (0x00)
and white (0xFF)
Binarization
Implementation (Simplest Approach) :
 Average the color as a cut off
0xFF
0x00
Slant
Difficulty :
 Different slant style @
ppl
Solution :
 Estimate the slant by
a slope
 Skew the image back
Slant Correction (Linear Regression)
Thinning
Purpose :
 Transform a binary image into one pixel
thickness
Thinning
Implementation (Simple Approach) :
 http://fourier.eng.hmc.edu/e161/lectures
/morphology/node2.html
Thinning

Will add more on implementation
Image Scaling


Image scaling is a process of resizing an image.
Mapping the pixels from the original image to the
destination image.
It’s non-trivial process that involves a trade-off
between efficiency, smoothness and sharpness.
Image Scaling

Scaling method used in this prototype
1st: Edge Detection:
2nd: Calculate scaler.
3rd: Do scaling: Convert scaler into integer,
enlarge every pixel by scaler times.
i.e. Scaler = 2.
Image Scaling

Pros: Easy to implement.

Cons: Suitable for integer scaler, not so
good for rational scaler.
Recognition
Usually machine learning algorithms
 Simple pixel-by-pixel difference would be
used as not closely related to this course.
 Will be implemented in later phase.

Gantt Chart
Program Live Demo
Q &A
The End
Thank you for you attentions!!