• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.95-98
• /
• 2002

문자인식 또는 음성인식을 위해 사용되어지는 CHMM(Continuous Hidden Markov Model)모델은 일반적으로 모델의 상태수를 일정한 수로 고정하는 고정 상태수 모델 구조를 가지고 있으나, 이는 개별적인 인식 단위의 특성을 고려하지 않은 경우로써 이를 고려한 가변 상태수 모델을 사용할 경우 인식률 향상을 기대할 수 있다. 개별적인 인식 단위에 적합한 모델 상태수를 결정하는 방법으로 파라미터 히스토그램 방법과, BIC(Bayesian Information Criterion)방법을 사용하는 것이 대표적이다. 이들 방법들은 개별적인 인식단위의 우도값만을 향상시키기 위한 방법으로 전체인식률과 직접적으로 비례하지는 않는다. 따라서, 본 논문에서는 고정 상태수를 갖는 모델 적용 방법과 인식단위별 상태수 변화에 따른 인식률을 비교하였으며, 이를 바탕으로 각 모델별 상태수를 달리하는 가변 상태수 CHMM모델 구성 방법을 제안한다. 제안된 가변상태수 모델의 유효성을 확인하기 위해 음성/문자 공용인식기 중 필기체 문자 인식에 적용한 결과 제안한 LM(Local Maximum)으로 구성된 가변 상태수 모델이 MLE와 BIC로 구성된 모델과 인식률 면에서는 거의 동일한 성능을 유지하면서 전체 상태수는 MLE 모델에 비해 $31\%$, BIC로 구성된 모델에 비해 $22\%$ 감소를 나타내어 제안한 모델의 유효성을 확인할 수 있었다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.439-441
• /
• 2019

Deep learning has become a powerful and robust algorithm in Artificial Intelligence. One of the most impressive forms of Deep learning tools is that of the Convolutional Neural Networks (CNN). CNN is a state-of-the-art solution for object recognition. For instance when we utilize CNN with MNIST handwritten digital dataset, mostly the result is well. Because, in MNIST dataset, all digits are centralized. Unfortunately, the real world is different from our imagination. If digits are shifted from the center, it becomes a big issue for CNN to recognize and provide result like before. To solve that issue, we have created frequency images from spatial images by a Fast Fourier Transform (FFT).

• The KIPS Transactions:PartB
• /
• v.10B no.4
• /
• pp.429-442
• /
• 2003

In this paper, we describe an address image recognition system for automatic processing of standard- size letter mail. The inputs to the system are gray-level mail piece images and the outputs are delivery point codes with which a delivery sequence of carrier can be generated. The system includes five main modules; destination address block location, text line separation, character segmentation, character recognition and finally address interpretation. The destination address block is extracted on the basis of experimental knowledge and the line separation and character segmentation is done through the analysis of connected components and vortical runs. For recognizing characters, we developed MLP-based recognizers and dynamical programming technique for interpretation. Since each module has been implemented in an independent way, the system has a benefit that the optimization of each module is relatively easy. We have done the experiment with live mail piece images directly sampled from mail sorting machine in Yuseong post office. The experimental results prove the feasibility of our system.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.26 no.1
• /
• pp.56-63
• /
• 2016

In this study, we introduce a design of Fuzzy RBFNNs-based digit recognition system using the incremental-PCA in order to recognize the handwritten digits. The Principal Component Analysis (PCA) is a widely-adopted dimensional reduction algorithm, but it needs high computing overhead for feature extraction in case of using high dimensional images or a large amount of training data. To alleviate such problem, the incremental-PCA is proposed for the computationally efficient processing as well as the incremental learning of high dimensional data in the feature extraction stage. The architecture of Fuzzy Radial Basis Function Neural Networks (RBFNN) consists of three functional modules such as condition, conclusion, and inference part. In the condition part, the input space is partitioned with the use of fuzzy clustering realized by means of the Fuzzy C-Means (FCM) algorithm. Also, it is used instead of gaussian function to consider the characteristic of input data. In the conclusion part, connection weights are used as the extended diverse types in polynomial expression such as constant, linear, quadratic and modified quadratic. Experimental results conducted on the benchmarking MNIST handwritten digit database demonstrate the effectiveness and efficiency of the proposed digit recognition system when compared with other studies.

• Annual Conference on Human and Language Technology
• /
• 1995.10a
• /
• pp.131-136
• /
• 1995

최근들어 자연스럽게 필기된 한글을 인식함으로써 정보 입력 과정을 자동화하기 위한 오프라인 한글 글씨 인식에 관한 연구가 활발히 진행되고 있다. 오프라인 한글 글씨 인식에 관한 연구에 있어서 반드시 확보되어야 하는 연구 환경으로 대용량 오프라인 한글 글씨 데이타베이스의 구축을 들 수 있는데, 본 논문에서는 시스템공학연구소 국어공학센터의 국어 정보 베이스 개발사업의 일환으로 추진중인 오프라인 한글 글씨 데이타베이스의 구축현황에 대해 간략히 소개하고자 한다. 오프라인 한글 글씨 데이타베이스의 구축은 크게 글씨 데이타베이스 설계, 글씨 데이타 수집, 용지 스캔 및 문자 단위 분할, 데이타베이스 검증의 4 단계로 구성된다. 본 연구에서는 다양한 변형을 갖는 글씨체의 수집을 데이타베이스 구축시 가장 고려해야 할 요소로 삼았으며, 고품질의 일관성 있는 글씨 데이타베이스 구축을 위해 데이타베이스 설계 단계와 검증 단계에 많은 시간을 할애했다. 마지막으로 본 연구에서는 WWW(World Wide Web)의 HTML(Hyper Text Markup Language)을 이용하여 편리 한 사용자 인터페이스를 구현함으로써 사용자들이 쉽게 한글 글씨 영상을 검색 할 수 있음은 물론 인식 알고리즘의 개발에 사용 가능한 형태의 화일을 제공받을 수 있도록 구성하고 있다. 현재는 KS C 완성형 한글 2,350자 중에서 사용 빈도순 상위 520자에 대한 한글 글씨 1,000벌을 수집하여 명도영상 데이타베이스를 구축 중에 있으며, 향후 2년간 나머지 1,830자에 대한 한글 글씨 데이타를 수집하여 데이타베이스를 완성하고자 한다. 구축된 글씨 데이타베이스는 조만간 국내의 오프라인 한글 글씨 인식 연구자들에게 제공되어 우수한 인식 알고리즘의 개발을 위한 중요한 실험 데이타로서 사용될 예정이며, 개발된 인식 시스템에 대한 객관적인 성능 평가에 있어서도 크게 기여하여 국내의 오프라인 한글 글씨 인식에 관한 연구를 활성화시켜주는 계기가 될 것으로 기대된다.

• The Transactions of the Korea Information Processing Society
• /
• v.3 no.5
• /
• pp.1281-1293
• /
• 1996

In this paper, we propose a novel recognition model of on-line cursive Korean characters using HMM(Hidden Markov Model) and level building algorithm. The model is constructed as a form of recognition network with HMM for graphemes and Korean combination rules. Though the network is so flexible as to accomodate variability of input patterns, it has a problem of recognition speed caused by 11, 172 search paths. To settle the problem, we modify the level building algorithm to be adapted directly to the Korean combination rules and apply it to the model. The modified algorithm is efficient network search procedure time complexity of which depends on the number of HMMs for each grapheme, not the number of paths in the extensive recognition network. A test with 15, 000 hand written characters shows recognition rat 90% and speed of 0.72 second per character.

• Journal of the Institute of Convergence Signal Processing
• /
• v.1 no.1
• /
• pp.64-73
• /
• 2000

The Korean character recognition technology for user interface in multimedia terminal requires fast processing time and high recognition rate. In this paper, we propose an phoneme and character recognition technology which uses characteristic information of korean and features of input strokes, i.e, feature point, feature vector, virtual vector, position relation between strokes. And, a recognition both phoneme and character by the various writing types of users uses korean database. The Korean database has been constructed by the characteristic information of korean and phoneme models which have various stroke information. Also, we use successive processing by the position relation between strokes and backtracking processing by the modification processing of stroke numbers which composed of each phoneme. This method reduces the complex processing of phoneme separation. The proposed on-line korean character recognition system has obtained 13msec average character processing time and correct recognition rate more than $95{\%}$ In a recognition experiment, where we tested 600 characters written by 10 people among 1,200 words.

• Journal of KIISE:Software and Applications
• /
• v.28 no.9
• /
• pp.681-687
• /
• 2001

In order to raise a class discrimination power by combining multiple classifiers under the Bayesian decision theory, the upper bound of a Bayes error rate bounded by the conditional entropy of a class variable and decision variables obtained from training data samples should be minimized. Wang and Wong proposed a tree dependence first-order approximation scheme of a high order probability distribution composed of the class and multiple feature pattern variables for minimizing the upper bound of the Bayes error rate. This paper presents an extended high order product approximation scheme dealing with higher order dependency more than the first-order tree dependence, based on the minimization of the upper bound of the Bayes error rate. Multiple recognizers for unconstrained handwritten numerals from CENPARMI were combined by the proposed approximation scheme using the Bayesian formalism, and the high recognition rates were obtained by them.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.8
• /
• pp.2400-2410
• /
• 2000

Many researches on recognition of Korean characters have been undertaken. But while the majority are done on Korean character recognition, tasks for developing document recognition system have seldom been challenged. In this paper, I designed a recognizer of Korean courtesy amounts to improve error correction in recognized character string. From the very first step of Korean character recognition, we face the enormous scale of data. We have 2350 characters in Korean. Almost the previous researches tried to recognize about 1000 frequently-used characters, but the recognition rates show under 80%. Therefore using these kinds of recognizers is not efficient, so we designed a statistical multiple recognizer which recognize 16 Korean characters used in courtesy amounts. By using multiple recognizer, we can prevent an increase of errors. For the Postprocessor of Korean courtesy amounts, we use the properties of Korean character strings. There are syntactic rules in character strings of Korean courtesy amounts. By using this property, we can correct errors in Korean courtesy amounts. This kind of error correction is restricted only to the Korean characters representing the unit of the amounts. The first candidate of Korean character recognizer show !!i.49% of recognition rate and up to the fourth candidate show 99.72%. For Korean character string which is postprocessed, recognizer of Korean courtesy amounts show 96.42% of reliability. In this paper, we suggest a method to improve the reliability of Korean courtesy amounts recognition by using the Korean character recognizer which recognize limited numbers of characters and the postprocessor which correct the errors in Korean character strings.

• Journal of Korea Multimedia Society
• /
• v.8 no.2
• /
• pp.181-191
• /
• 2005

The classification error of 'ㅁ', 'ㅇ' is one of the main causes of incorrect recognition in Korean characters, but there haven't been enough researches to solve this problem. In this paper, a new feature extraction method from Korean grapheme is proposed to recognize 'ㅁ', 'ㅇ'effectively. First, we defined an optimal neighbor-distance selection measure using modified Hausdorff distance, which we determined the optimal neighbor-distance by. And we extracted neighbor-angle feature which was used as the effective feature to classify the two graphemes 'ㅁ', 'ㅇ'. Experimental results show that the proposed feature extraction method worked efficiently with the small number of features and could recognize the untrained patterns better than the conventional methods. It proves that the proposed method has a generality and stability for pattern recognition.