• Journal of the Korean Institute of Intelligent Systems
• /
• v.25 no.4
• /
• pp.355-360
• /
• 2015

In this paper, we propose the design of Radial Basis Function Neural Network based on PCA in order to recognize handwritten digits. The proposed pattern classifier consists of the preprocessing step of PCA and the pattern classification step of pRBFNNs. In the preprocessing step, Feature data is obtained through preprocessing step of PCA for minimizing the information loss of given data and then this data is used as input data to pRBFNNs. The hidden layer of the proposed classifier is built up by Fuzzy C-Means(FCM) clustering algorithm and the connection weights are defined as linear polynomial function. In the output layer, polynomial parameters are obtained by using Least Square Estimation (LSE). MNIST database known as one of the benchmark handwritten dataset is applied for the performance evaluation of the proposed classifier. The experimental results of the proposed system are compared with other existing classifiers.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.10 no.5
• /
• pp.979-984
• /
• 2009

In this paper, the technique for production and encoding of new dot-type print watermark is proposed. The print watermark has characteristics similar to those of the digital watermark and function as link which change various first informations(texts, symbols, figures, etc) on the printed matter to secondary contents (sound, video, character, etc) corresponding each to informations on the printed matter. The proposed dot-type print watermark pattern is represented as $16{\times}16$ matrix in $0.4mm^2$ area and dots are printed on only 23 elements of $16{\times}16$ matrix. The size of each dot is so small(0.02mm)that it can not be seen. Because position of printed dots correspond to the position of each digit in binary notation in this paper, they are encoded easily and there are about 8,000,000 watermark patterns enough to express first information of printed matters. It was showed that the proposed print watermark patterns are recognized without difficulty by the own recognition device.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.12 no.3
• /
• pp.201-207
• /
• 2012

A RAM-based neural network is a weightless neural network based on binary neural network(BNN) which is efficient neural network with a one-shot learning. RAM-based neural network has multiful information bits and store counts of training in BNN. Supervised learning based on the RAM-based neural network has the excellent performance in pattern recognition but in pattern categorization with unsupervised learning as unsuitable. In this paper, we propose a unsupervised learning algorithm in the RAM-based neural network to perform pattern categorization. By the proposed unsupervised learning algorithm, RAM-based neural network create categories depending on the input pattern by itself. Therefore, RAM-based neural network for supervised learning and unsupervised learning should proof of all possible complex models. The training data for experiments provided by the MNIST offline handwritten digits which is consist of 0 to 9 multi-pattern.

• 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 KIISE:Software and Applications
• /
• v.28 no.12
• /
• pp.875-884
• /
• 2001

Although the self organizing map (SOM) is widely utilized in such fields of data visualization and topology preserving mapping, since it should have the topology fixed before trained, it has some shortcomings that it is difficult to apply it to practical problems, and classification capability is quite low despite better clustering performance. To overcome these points this paper proposes the dynamic topology preserving self-organizing map(DTSOM) that dynamically splits the output nodes on the map and trains them, and attempts to improve the classification capability by combining multiple DTSOMs K-Winner method has been applied to combine DTSOMs which produces K outputs with winner node selection method. This produces even better performance than the conventional combining methods such as majority voting weighting, BKS Bayesian, Borda, Condorect and reliability sum. DTSOM remedies the shortcoming of determining the topology in advance, and the classification rate increases significantly by combing multiple maps trained with different features. Experimental results with handwritten digit recognition indicate that the proposed method works out to problems of conventional SOM effectively so to improve the classification rate to 98.1%.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.11 no.3
• /
• pp.223-230
• /
• 2001

Since self-organizing map (SOM) preserves the topology of ordering in input spaces and trains itself by unsupervised algorithm, it is Llsed in many areas. However, SOM has a shortcoming: structure cannot be easily detcrmined without many trials-and-errors. Structure-adaptive self-orgnizing map (SASOM) which can adapt its structure as well as its weights overcome the shortcoming of self-organizing map: SASOM makes use of structure adaptation capability to place the nodes of prototype vectors into the pattern space accurately so as to make the decision boundmies as close to the class boundaries as possible. In this scheme, the initialization of weights of newly adapted nodes is important. This paper proposes a method which optimizes SASOM with genetic algorithm (GA) to determines the weight vector of newly split node. The leanling algorithm is a hybrid of unsupervised learning method and supervised learning method using LVQ algorithm. This proposed method not only shows higher performance than SASOM in terms of recognition rate and variation, but also preserves the topological order of input patterns well. Experiments with 2D pattern space data and handwritten digit database show that the proposed method is promising.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.7 no.1
• /
• pp.59-67
• /
• 2008

In this paper we propose an algorithm to extract the license plate regions from Korean car images. The idea of this paper is that we first find the four digits in the input car image and then segment the plate region using the digit information. Out method has advantage of segmenting simultaneously the plate regions and four digits regions. The first step finds and groups the connected components with proper sizes as candidate digits. The second step applies an serial alignment condition to find out probable 4-digits. In the third step, we recognize the candidate digits and assign the confidence values to each of them. The final step extracts the license plate region which has the highest confidence value. We used the Perfect Metrics classification algorithm to estimate the confidence. In our experiment, we got 97.23% and 95.45% correct detection rates, 0.09% and 0.11% false detection rates for 4,600 daytime and 264 nighttime images, respectively.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.27 no.5
• /
• pp.1033-1047
• /
• 2017

Traditional explicit authentication, which requires awareness of the user's authentication process, is a burden on the user, which is one of main reasons why users tend not to employ authentication. In this paper, we try to reduce such cost by employing implicit authentication methods, such as biometrics and location based authentication methods. We define the 4-level security assurance model, where each level is mapped to an explicit authentication method. We implement our model as an Android application, where the implicit authentication methods are touch-stroke dynamics-based, face recognition based, and the location based authentication. From user experiment, we could show that the authentication cost is reduced by 14.9% compared to password authentication-only case and by 21.7% compared to the case where 6-digit PIN authentication is solely used.

• The Transactions of the Korea Information Processing Society
• /
• v.3 no.3
• /
• pp.449-459
• /
• 1996

The main goal of this paper is to introduce a novel definition for fuzzy mathematical morphology and a neural network implementation. The generalized- mean operator plays the key role for the definition. Such definition is well suited for neural network implementation. The first stage of the shared-weight neural network has adequate architecture to perform morphological operation. The shared- weight network performs classification based on the features extracted with the fuzzy morphological operation defined in this paper. Therefore, the parameters for the fuzzy definition can be optimized using neural network learning paradigm. Learning rules for the structuring elements, degree of membership, and weighting factors are precisely described. In application to handwritten digit recognition problem, the fuzzy morphological shared-weight neural network produced the results which are comparable to the state-of art for this problem.

• Journal of the Korea Society of Computer and Information
• /
• v.14 no.8
• /
• pp.11-18
• /
• 2009

This paper proposes the way of improving learning speed in Levenberg-Marquardt algorithm using the principal submatrix of Jacobian matrix. The Levenberg-Marquardt learning uses Jacobian matrix for Hessian matrix to get the second derivative of an error function. To make the Jacobian matrix an invertible matrix. the Levenberg-Marquardt learning must increase or decrease ${\mu}$ and recalculate the inverse matrix of the Jacobian matrix due to these changes of ${\mu}$. Therefore, to have the proper ${\mu}$, we create the principal submatrix of Jacobian matrix and set the ${\mu}$ as the eigenvalues sum of the principal submatrix. which can make learning speed improve without calculating an additional inverse matrix. We also showed that our method was able to improve learning speed in both a generalized XOR problem and a handwritten digit recognition problem.