• Annual Conference on Human and Language Technology
• /
• 1996.10a
• /
• pp.93-97
• /
• 1996

본 논문에서는 2 계층 분류기를 이용하여 일반적인 문서(보고서, 책, 잡지, 워드프로세서에서 출력 된 양식) 내의 다중 크기 및 다중 활자체의 인식을 위한 효과적인 방법을 제안하고 구현하였다. 다중언어 문자를 효과적으로 인식하기 위한 2 계층 분류기를 제안하였는데 이는 폰트 독립적 분류기와 폰트 의존적 분류기로 구성되어 있다. 제안된 방법의 성능 평가를 위하여 사무실에서 많이 사용하는 59 종류의 폰트와 각 폰트 당 3가지 크기의 글꼴과, 스캐너에서 지원되는 3가지 농도의 총 489개의 서로 다른 부류를 갖는 3,593,172 자를 대상으로 학습시킨 뒤에 일반 문서를 가지고 펜티엄 PC 상에서 인식 실험을 수행하였다. 실험 결과, 2계층 분류기를 갖는 시스템에서 96-98%의 인식률과 초당40자 이상의 인식 속도를 보여줌으로써 일반적인 문서에서 다중 크기 및 다중 활자체의 문자 인식에 매우 실용적인 가치가 있음을 확인했다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.17 no.11
• /
• pp.1175-1184
• /
• 1992

A new classification method and recognition algorithms for printed Korean and Chinese character is studied for Korean text which contains both Korean and Chinese characters. The proposed method utilizes structural features of the vertical and horizontal vowel in Korean character. Korean characters are classified into 6 groups. Vowel and consonant are separated by means of different vowel extraction methods applied to each group. Time consuming thinning process is excluded. A modified crossing distance feature is measured to recognize extracted consonant. For Chinese character, an average of stroke crossing number is calculated on every characters, which allows the characters to be classified into several groups. A recognition process is then followed in terms of the stroke crossing number and the black dot rate of character. Classification between Korean and Chinese character was at the rate of 90.5%, and classification rate of Ming-style 2512 Korean characters was 90.0%. The recognition algorithm was applied on 1278 characters. The recognition rate was 92.2%. The densest class after classification of 4585 Chinese characters was found to contain only 124 characters, only 1/40 of total numbers. The recognition rate was 89.2%.

• The KIPS Transactions:PartB
• /
• v.10B no.7
• /
• pp.777-784
• /
• 2003

In this Paper. we propose a recognition method of machine printed characters based on the combination of recognition units using multiple neural networks. In our recognition method, the input character is classified into one of 7 character types among which the first 6 types are for Hangul character and the last type is for non-Hangul characters. Hangul characters are recognized by several MLP (multilayer perceptron) neural networks through two stages. In the first stage, we divide Hangul character image into two or three recognition units (HRU : Hangul recognition unit) according to the combination fashion of graphemes. Each recognition unit composed of one or two graphemes is recognized by an MLP neural network with an input feature vector of pixel direction angles. In the second stage, the recognition aspect features of the HRU MLP recognizers in the first stage are extracted and forwarded to a subsequent MLP by which final recognition result is obtained. For the recognition of non-Hangul characters, a single MLP is employed. The recognition experiments had been performed on the character image database collected from 50,000 real letter envelope images. The experimental results have demonstrated the superiority of the proposed method.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.34 no.9B
• /
• pp.908-913
• /
• 2009

Mobile phones became important household appliance that cannot be without in our daily lives. And the short messaging service (SMS) in these mobile phones is 1.5 to 2 times more than the voice service. However, the spam filtering functions installed in mobile phones take a method to receive specific number patterns or words and recognize spam messages when those numbers or words are present. However, this method cannot properly filters various types of spam messages currently dispatched. This paper proposes a more powerful and more adaptive spam filtering system using SVM and thesaurus. The system went through a process of isolating words from sample data through pro-processing device and integrating meanings of isolated words using a thesaurus. Then it generated characteristics of integrated words through the chi-square statistics and studied the characteristics. The proposed system is realized in a Window environment and the performance is confirmed through experiments.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2000.10b
• /
• pp.927-930
• /
• 2000

본 논문은 문서 영상에 대해 투영을 사용하여 영역을 나누었고 각 영역에 대해 고주파 밴드의 웨이블렛 계수의 통계적 분산과 히스토그램을 기반으로 한 두 가지 특징을 사용하여 문자와 그림으로 분류하였다. 투영으로 나누어진 영역들에 대해 일정 크기의 블록으로 나누고 두 가지 특징에 따라 문자와 그림으로 분류하였다. 따라서 투영에 의해 나뉜 영역 중 문자와 그림이 혼합되어 의미가 모호한 영역에 대해 잘못 분류되는 가능성을 줄일 수 있었다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2017.01a
• /
• pp.157-160
• /
• 2017

본 연구에서는 한글 문자 인식을 수행하였다. 한글 인식을 수행하되 고딕 인쇄체 문자를 대상으로 하였고, 자소 단위 인식을 통한 인식을 수행하되 기존 한글 문자 인식 연구에서 사용하는 자음과 모음 단위의 자소가 아닌 연결성분을 이용하여 인식하는 새로운 자소를 이용하였다. 새로운 자소들은 끝점, 2선 모임점, 3선 모임점, 4선 모임점의 특징을 추출하고 특징에 의해 자소를 인식하는 데이터베이스를 구성하여 자소를 인식하게 하였다. 또한 연결 성분을 반영한 새로운 자소로 고딕 인쇄체 문자를 인식하므로 추출된 자소를 6가지로 분류하였고, 6가지 자소에 의해 구성되는 92가지 문자 구조를 제안하고 이에 따른 문자를 데이터베이스를 구축하였고, 자소의 무게 중심을 이용한 분포를 이용하여 제안된 구조를 통하여 데이터베이스를 이용한 문자인식을 수행하였다.

• Journal of Digital Convergence
• /
• v.10 no.11
• /
• pp.383-387
• /
• 2012

Development of automated document management system specified for scanned invoice images suffers from rigorous accuracy requirements for extraction of monetary data, which necessiate automatic validation on the extracted values for a generative invoice table model. Use of certain internal constraints such as "amount = unit price times quantity" is typical implementation. In this paper, we propose a noble invoice information extraction model with improved auto-validation method by utilizing table header detection and column classification.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06b
• /
• pp.306-308
• /
• 2012

핸드폰 사용이 보편화되고 핸드폰의 문자 사용량이 늘어감에 따라 대량의 핸드폰 문자 메시지를 구축하는 건이 가능해졌다. 이러한 문자 데이터를 처리에 기반이 되는 대화 구분 방법을 제안하였다. 이 방법론은 기존 문서분류 방식을 적용하는데 발생하는 문제를 피하기 위해 시간정보를 사용하는 비지도학습 방법론이다. 해당 방법을 실제 핸드폰 메시지 데이터에 적용한 결과 정확율과 재현율에서 0.9를 넘는 높은 성능을 보였다.

• Journal of KIISE:Software and Applications
• /
• v.27 no.6
• /
• pp.660-667
• /
• 2000

In this paper, a word segmentation method for handwritten Korean text line images is proposed. The method uses gap information to segment words in line images, where the gap is defined as a white run obtained after vertical projection of line images. Each gap is assigned to one of inter-word gap and inter-character gap based on gap distance. We take up three distance measures which have been proposed for the word segmentation of handwritten English text line images. Then we test three clustering techniques to detect the best combination of gap metrics and classification techniques for Korean text line images. The experiment has been done with 305 text line images extracted manually from live mail pieces. The experimental result demonstrates the superiority of BB(Bounding Box) distance measure and sequential clustering approach, in which the cumulative word segmentation accuracy up to the third hypothesis is 88.52%. Given a line image, the processing time is about 0.05 second.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.12
• /
• pp.61-67
• /
• 2021

In this paper, we propose a method of extracting and recognizing related information for research from images of the unstructured pulmonary function test papers using character detection and recognition techniques. Also, we develop a post-processing method to reduce the character recognition error rate. The proposed structuring method uses a character detection model for the pulmonary function test paper images to detect all characters in the test paper and passes the detected character image through the character recognition model to obtain a string. The obtained string is reviewed for validity using string matching and structuring is completed. We confirm that our proposed structuring system is a more efficient and stable method than the structuring method through manual work of professionals because our system's error rate is within about 1% and the processing speed per pulmonary function test paper is within 2 seconds.