KAIST Hanja DB1

Built by

Cognitive Systems Group, AI laboratory, Computer Science Department, KAIST, 1996

 

Data Collection

 

The images are written by little more than 200 peoples. The 800 most frequently used character classes in names of Korean, which covers 96.6% of usage, are chosen to be collected. The subjects wrote Chinese characters on sheets containing 800 fields each of which is for one character. The sheets are scanned with flatbed scanner, and segmented into characters.

The characters are filtered and sorted according to quality manually. Among 800 classes, 17 classes are discarded and 783 class are remained.  

 

Quality

• sample size : 783 classes * 200 samples, 279,514 Kbytes
• image quality : good
• resolution : 300 dpi binary images
• variation : normally written (not very cursive)

 

Database Structure

 

The database contains 783 files, each of which has 200 samples

 

Data Format

The data is stored in KS format designed by Samsung Electronic Company.

File is composed of file header and data array

File header is 4 bytes string 'KS  ' indicating KS format.

Data array is a sequence of character image, each of which is as follows.

 

Name	Size in bytes	Description
code	2	KS-5601 code
width	1	width of the image
height	1	height of the image
IsOK	1	0 means poor quality image 1 means good quality image
image	(width+7)/8*height	1bit/pixel, row-major coding

 

Examples

 

 

 

Download small size samples

• 1st samples - best quality (273 Kbytes)
• 100th samples - medium quality (244 Kbytes)
• 200th samples - worst quality(245 Kbytes)

Contact to

• Tel. :+82-42-869-5557
• Fax. : +82-42-869-3510
• E-mail : webmaster@ai.kaist.ac.kr