• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2018.05a
• /
• pp.693-695
• /
• 2018

In this paper, we propose the improvement of Hangul OCR accuracy through deep learning. OCR is a program that senses printed and handwritten characters in an optical way and encodes them digitally. In the case of the most commonly used Tesseract OCR, the accuracy of English recognition is high. However, Hangul has lower accuracy because it has less learning data for a complex structure. Therefore, in this study, we propose a method to improve the accuracy of Hangul OCR by extracting the character region from the desired image through image processing and using deep learning using it as learning data. It is expected that OCR, which has been developed only by existing alphanumeric and several languages, can be applied to various languages.

• Korean Journal of Cognitive Science
• /
• v.9 no.1
• /
• pp.1-11
• /
• 1998

In this paper, the features which are thought to be used by humans based on the psychological experiment of human information processing are applied to character recognition problem. Man will deal with a little large area information as well as pixel by pixel information. Therefore we define the feature that represents a little wide region I information called region feature, and combine the features derived from region feature and pixel by pixel features that have been used by now. The features we used are the result of region feature based preanalysis, mesh with region attributes, cross distance difference and gradient. The training and test data in the experiment are handwritten Korean alphabets, digits and English alphabets, which are trained on neural network using back propagation algorithm and recognition results are 90.27-93.25%, 98.00% and 79.73-85.75%, respectively Experimental results show that the feature we are suggesting in this paper is 1-2% better than UDLRH feature similar in attribute to region feature, and the tendency of misrecognition is more easily acceptable by humans.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.43 no.5 s.311
• /
• pp.30-36
• /
• 2006

In this paper, a new type of written image segmentation based on relative comparison of region areas is proposed. The original image is composed of two distinctive regions; information and background. Compared with this binary original image, the observed one is the gray scale which is represented with complex regions with speckles and noise due to degradation or contamination. For applying threshold or statistical approach, there occurs the region-deformation problem in the process of binarization. At first step, the efficient iterated conditional mode (ICM) which takes the lozenge type block is used for regions formation into the binary image. Secondly the information region is estimated through selecting action and restored its primary state. Not only decision of the attachment to a region but also the calculation of the magnitude of its area are carried on at each current pixel iteratively. All region areas are sorted into a set and selected through the decision parameter which is obtained statistically. Our experiments show that these approaches are effective on ink-rubbed copy image (拓本 'Takbon') and efficient at shape restoration. Experiments on gray scale image show promising shape extraction results, comparing with the threshold-segmentation and conventional ICM method.

• Journal of the Korea Society of Computer and Information
• /
• v.15 no.10
• /
• pp.49-57
• /
• 2010

In this thesis, We studied on hand-written character recognition, that characters entered into a digital input device and remove noise and separating character elements using preprocessing. And processed character images has done thinning and 3-level wavelet transform for making normalized image and reducing image data. The structural method among the numerical Hangul recognition methods are suitable for recognition of printed or hand-written characters because it is usefull method deal with distortion. so that method are applied to separating elements and analysing texture. The results show that recognition by analysing texture is easily distinguished with respect to consonants. But hand-written characters are tend to decreasing successful recognition rate for the difficulty of extraction process of the starting point, of interconnection of each elements, of mis-recognition from vanishing at the thinning process, and complexity of character combinations. Some characters associated with the separation process is more complicated and sometime impossible to separating elements. However, analysis texture of the proposed character recognition with the exception of the complex handwritten is aware of the character.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.7 no.5
• /
• pp.1044-1051
• /
• 2003

This paper surveys the EBP(Error Back Propagation) learning, the Cross Entropy function and the LBL(Layer By Layer) learning, which are used for learning the MLP(Multi Layer Perceptrons). We compare the merits and demerits of each learning method in the handwritten digit recognition. Although the speed of EBP learning is slower than other learning methods in the initial learning process, its generalization capability is better. Also, the speed of Cross Entropy function that makes up for the weak points of EBP learning is faster than that of EBP learning. But its generalization capability is worse because the error signal of the output layer trains the target vector linearly. The speed of LBL learning is the fastest speed among the other learning methods in the initial learning process. However, it can't train for more after a certain time, it has the lowest generalization capability. Therefore, this paper proposes the standard of selecting the learning method when we apply the MLP.

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.27 no.10
• /
• pp.68-77
• /
• 1990

In this paper, a neural patten recognition method for the automatic circuit diagram reading system is proposed. The proposed procedure to recognize a deformed logic symbols is composed of three stages: feature detection, log mapping, and pattern classification. In the feature detection stage, a modified competitive learning algorithm where each pattern has the inhibition weight as well as the activation weight is developed. The global information of hand-written logic symbols is obtained by the feature detection neural network having both the inhibition and activation weights. The obtained global data is then transformed into a log space by the conformal mapping where according to the Schwartz's theory about the human visual signal process-ing, the degree of rotation and the scale change are mapped into the translation change. Logic symbols are finally classified by a three layer perceptron trained by the error back propagation algorithm. The computer simulation demonstrates that the proposed multistage neural network system can recognize well the deformed patterns of hand-written logic circuit diagrams.

• Journal of Digital Convergence
• /
• v.16 no.2
• /
• pp.9-17
• /
• 2018

The purpose of this study is to find out how voice recognition technology can be utilized to solve the elderly problem rapidly aging in Korea. Public support services and civilian nursing services for the elderly are expected to expand in Korea. In this case, voice recognition technology can be used variously for the elderly who are not familiar with the media interface. To this end, our researchers visited Japan and examined the achievements obtained by voice recognition technology in the elderly care. Especially, when caregivers write reports, they have greatly reduced their working hours by replacing the handwritten reports with ones using voice recognition technology. This method can be easily implemented in Korea. In addition, the social cost of the elderly support can be gradually reduced through the development of a robot equipped with voice recognition technology. Consequently, we realize that when voice recognition technology is combined with artificial intelligence programs of various emotion recognition functions and various policy possibilities as well.

• Korean Journal of Cognitive Science
• /
• v.7 no.1
• /
• pp.75-84
• /
• 1996

In this paper we emphasize the importance of architectural aspects of designing a handwritten numeral recognition program. and describe two architectural design.First, we describe the modular design of a numeral recognition program, and mention its advantages.In this design, a recognizer is composed of 10 binary subrecognizers each of which is responsible for only one class.Rule-based training and neural-based training are presented.Second, we connect two(or more)recognizers serially which we call pipelining connection.The second recognizer may act as verifier for the patterns recognized by the forst recognizer, or as second chance recognizer for the patterns rejected by the first recognizer.Our experimental results obtained till now show the merits of the proposed architectural designs.

• 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.

• Journal of KIISE:Software and Applications
• /
• v.28 no.9
• /
• pp.603-611
• /
• 2001

This paper proposes a novel fast layer-by-layer algorithm that has better generalization capability. In the proposed algorithm, the weights of the hidden layer are updated by the target vector of the hidden layer obtained by least squares method. The proposed algorithm improves the learning speed that can occur due to the small magnitude of the gradient vector in the hidden layer. This algorithm was tested in a handwritten digits recognition problem. The learning speed of the proposed algorithm was faster than those of error back propagation algorithm and modified error function algorithm, and similar to those of Ooyen's method and layer-by-layer algorithm. Moreover, the simulation results showed that the proposed algorithm had the best generalization capability among them regardless of the number of hidden nodes. The proposed algorithm has the advantages of the learning speed of layer-by-layer algorithm and the generalization capability of error back propagation algorithm and modified error function algorithm.