NEIS 평가시스템 기능 향상을 위한 필기체 한글 인식

Metadata Downloads
Issued Date
Two types of tests are widely used for multilateral assessment. One is the regular test, which is usually held twice a semester; a midterm and a final. The other is the occasional one, called Performance Test, which is meant to complement the former one. Still, the regular one, which is usually multiple choice type is the major criterion for the assessment of students' growth. Evaluation process equipments such as OMR cards and OMR card readers have made the evaluation process simple and speedy, and that is the reason why the multiple choice type test is much preferred.
The objective test has brought about quite a few side-effects so far. The cell phone cheating on CSAT(College Scholastic Aptitude Test) last year revealed the vulnerability of the multiple choice type test. That scandal was the biggest social event in 2004, and still remains to be a national headache. With the objective test items, students are asked to choose the one right answer, slowly and unconsciously becoming narrow-minded. Choosing the most probable one out of five sometimes means that they can be lucky to win the odds for the bet, exhilarated by the accidental profit. The result itself has been laid before the process. It is very hard to cover all the contents taught with the limited choice items. Furthermore, asking only discrete knowledge, the objective test may be blocking students' thinking process or paralyzing problem solving ability.
To solve the problems mentioned above, Neural Network, which was modeled after the human brain, is now being used. Neural Network was believed to be a promising substitute for Von Neumann's computing model. When it comes to the Korean handwriting recognition, a neural network trained by Backpropagation method is being used in many researches.
This research aims to develop the Korean handwriting recognition method with the help of Backpropagation Neural Network Learning Method, apply the method to the evaluation system, and improve the problems derived from the multiple choice type test.
With the present OMR cards and OMR card readers, only multiple choice items can be processed technically. Thus, a new type of OMR card that can process the short answer test as well as the multiple choice test is suggested here. This card can be used widely regardless of the personal information and the subject. Also, it enables students to write both the subjective items and the objective items on the same side so that the scoring process can be easily mechanized. On the A4-sized answer card, objective answer part becomes more spacious, allowing students to mark the answers with more ease. As for the short answers, up to 26 syllables can be written into the answer cell.
With the new answer card, objective answers were scored by scanning the written images, instead of using the OMR card readers. Objective answer part was regularly divided into smaller parts according to the number of choices, the number of the black pixels within each part was counted, and the part with the most pixels was recognized as the student's answer.
When the short written answers were scored, every syllable was divided into graphemes and strokes were extracted. Extracted strokes were presented according to the probability of their appearance, and then trained with six Type Recognition Neural Network. They were trained and recognized again with the specific neural network. After trained by PE92, the certified DB, and recognized by using the result of the subjective answers, the recognition ratio was 79.2%. This ratio is higher than the ratio in previous research.
Most part of NEIS(National Education Information System) is adopting the objective evaluation method. To apply the subjective evaluation method suggested above, the evaluation process part of NEIS has been reconstructed here. With the reconstructed system, teachers can search and correct the grades without any help from the teacher in charge of evaluation process part. Subjective answer part, which used to be marked only with the whole grades, was revised so that teachers can input the marks into each item.
Experiments with the method suggested in this paper showed that the degree of discrimination between students' achievement was improved and learning-teaching method was also promoted. Above all, mechanization of the scoring the subjective answers has brought much convenience to teachers.
Alternative Title
Handwriting Hangul Recognition for Improvement of Evaluation System on NEIS
Alternative Author(s)
Kim, Yong-Hun
조선대학교 대학원
일반대학원 컴퓨터공학
Awarded Date
Table Of Contents
제 1 장 서론 = 1
제 1 절 연구배경 = 1
제 2 절 연구목표 = 3
제 2 장 관련연구 = 5
제 1 절 한글의 계층적 분해 = 5
제 2 절 유형 분류를 위한 신경망 선택 = 12
제 3 절 오류역전파(BP) 알고리즘 = 13
1. 델타 학습법 = 15
2. 일반 델타 학습법 = 21
3. BP 알고리즘 = 25
4. 학습 인자 = 29
제 3 장 평가시스템 분석 = 32
제 1 절 평가의 개념 = 32
1. 최근의 교육평가 = 32
2. 교육평가의 유형 = 33
3. 학교에서의 평가의 절차 = 34
제 2 절 나이스(NEIS)시스템 = 37
1. 나이스(NEIS)시스템의 소개 = 37
2. 나이스 성적처리 순서 = 39
제 3 절 바람직한 평가 방법 = 50
제 4 절 OMR 카드 설계 = 52
제 5 절 성적 처리 시스템 재구성 = 54
제 4 장 필기체 한글 인식 = 57
제 1 절 전처리 = 57
1. 히스토그램 = 57
2. 히스토그램 균일화 = 58
3. 세선화 = 59
제 2 절 객관식 처리 = 62
제 3 절 주관식 처리 = 67
1. 관련 연구 = 67
2. 필기체 한글인식 = 68
3. 제안된 방법을 이용한 특징 추출 = 73
4. 실험 결과 = 73
제 4 절 인식 결과 = 87
제 5 장 NEIS 평가시스템에 적용 및 결과 = 90
제 6 장 결론 = 94
참고문헌 = 96
조선대학교 대학원
김용훈. (2006). NEIS 평가시스템 기능 향상을 위한 필기체 한글 인식.
Appears in Collections:
General Graduate School > 4. Theses(Ph.D)
Authorize & License
  • AuthorizeOpen
  • Embargo2008-06-27
Files in This Item:

Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.