• Journal of the Korean Institute of Telematics and Electronics
• /
• v.27 no.11
• /
• pp.144-151
• /
• 1990

In this work Ohotomo et al., neural network model for learning and recognizing vowels is modified in order to reduce the time for learning and the possibility of incorrect recognition. In this modification, the finite bandwidth of formant frequencies of vowels are taken into consider-ations in coding input patterns. Computer simulations show that the modification reduces not only the possibility of incorrect recognition by about $30{\%}$ but also the time for learning by about $7{\%}$.

• The KIPS Transactions:PartB
• /
• v.15B no.1
• /
• pp.53-60
• /
• 2008

This paper proposes a new multiobjective optimization method for discriminative training of hidden Markov models (HMMs) used as the recognizer for automatic lipreading. While the conventional Baum-Welch algorithm for training HMMs aims at maximizing the probability of the data of a class from the corresponding HMM, we define a new training criterion composed of two minimization objectives and develop a global optimization method of the criterion based on simulated annealing. The result of a speaker-dependent recognition experiment shows that the proposed method improves performance by the relative error reduction rate of about 8% in comparison to the Baum-Welch algorithm.

• The KIPS Transactions:PartB
• /
• v.15B no.3
• /
• pp.229-236
• /
• 2008

This paper examines effectiveness of using color images instead of grayscale ones for automatic lipreading. First, we show the effect of color information for performance of humans' lipreading. Then, we compare the performance of automatic lipreading using features obtained by applying principal component analysis to grayscale and color images. From the experiments for various color representations, it is shown that color information is useful for improving performance of automatic lipreading; the best performance is obtained by using the RGB color components, where the average relative error reductions for clean and noisy conditions are 4.7% and 13.0%, respectively.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.36C no.11
• /
• pp.75-83
• /
• 1999

We propose a PfSGA(parameter-free species genetic algorithm) to learn the topological structure of MLP classifiers being adequate to given applications. The PfSGA is a combinational method of SGA(species genetic algorithm) and PfGA(parameter-free genetic algorithm). In SGA, we divide the total search space into several subspaces(species) according to the number of hidden units, and reduce the unnecessary search by eliminating the low promising species from the evolutionary process. However the performances of SGA classifiers are readily affected by the values of parameters such as mutation ratio and crossover ratio. In this paper, therefore, we combine SGA with PfGA, for which it is not necessary to determine the learning parameters. Experimental results on benchmark data and sign language words show that PfSGA can reduce the learning time of SGA and is not affected by the selection parameter values on structural learning. The results also show that PfSGA is more efficient than the exisiting methods in the aspect of misclassification ratio, learning rate, and complexity of MLP structure.

• The Journal of the Acoustical Society of Korea
• /
• v.19 no.5
• /
• pp.20-26
• /
• 2000

In speaker verification system the HMM(hidden Markov model) parameter updating using small amount of data and the priori threshold decision are crucial factor for dealing with long-term variability in people voices. In the paper we present the speaker model updating technique which can be adaptable to the session-to-intra speaker variability and the priori threshold determining technique. The proposed technique decreases verification error rates which the session-to-session intra-speaker variability can bring by adapting new speech data to speaker model parameter through Baum Welch re-estimation. And in this study the proposed priori threshold determining technique is decided by a hybrid score measurement which combines the world model based technique and the cohen model based technique together. The results show that the proposed technique can lead a better performance and the difference of performance is small between the posteriori threshold decision based approach and the proposed priori threshold decision based approach.

• Journal of the Korea Society of Computer and Information
• /
• v.10 no.4 s.36
• /
• pp.203-211
• /
• 2005

In this paper, we propose a new automatic fingerprint identification system that identifies individuals in large databases. The algorithm consists of three steps; preprocessing, classification, and matching, in the classification. we present a new classification technique based on the statistical approach for directional image distribution. In matching, we also describe improved minutiae candidate pair extraction algorithm that is faster and more accurate than existing algorithm. In matching stage, we extract fingerprint minutiaes from its thinned image for accuracy, and introduce matching process using minutiae linking information. Introduction of linking information into the minutiae matching process is a simple but accurate way, which solves the problem of reference minutiae pair selection in comparison stage of two fingerprints quickly. This algorithm is invariant to translation and rotation of fingerprint. The proposed system was tested on 1000 fingerprint images from the semiconductor chip style scanner. Experimental results reveal false acceptance rate is decreased and genuine acceptance rate is increased than existing method.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.9 no.5
• /
• pp.543-548
• /
• 2014

Multiple-model based speech recognition framework (MMSR) has been known to be very successful in speech recognition. Since it uses multiple hidden Markov modes (HMMs) that corresponds to various noise types and signal-to-noise ratio (SNR) values, the selected acoustic model can have a close match with the test noisy speech. However, since the number of HMM sets is limited in practical use, the acoustic mismatch still remains as a problem. In this study, we experimentally determined the optimal SNR mapping between the test noisy speech and the HMM set to mitigate the mismatch between them. Improved performance was obtained by employing the SNR mapping instead of using the estimated SNR from the test noisy speech. When we applied the proposed method to the MMSR, the experimental results on the Aurora 2 database show that the relative word error rate reduction of 6.3% and 9.4% was achieved compared to a conventional MMSR and multi-condition training (MTR), respectively.

• The KIPS Transactions:PartB
• /
• v.10B no.7
• /
• pp.815-822
• /
• 2003

Fingerprint image enhancement and minutiae matching are two key steps in an automatic fingerprint identification system. In this paper, we propose a fingerprint recognition technique by using minutiae linking information. Recognition process have three steps ; preprocessing, minutiae extraction, matching step based on minutiae pairing. After extracting minutiae of a fingerprint from its thinned image for accuracy, we introduce matching process using minutiae linking information. Introduction of linking information into the minutiae matching process is a simple but accurate way, which solves the problem of reference minutiae pair selection with low cost in comparison stage of two fingerprints. This algorithm is invariable to translation and rotation of fingerprint. The matching algorithm was tested on 500 images from the semiconductor chip style scanner, experimental result revealed the false acceptance rate is decreased and genuine acceptance rate is increased than existing method.

• Journal of the Korea Society of Computer and Information
• /
• v.14 no.7
• /
• pp.73-80
• /
• 2009

Internet accidents have skyrocketed every year. It always has been threatened by the methods such as hacking and Spyware. The majority of security accident is formed of the loss of authentication information, and the internal user who is not authorized. The importance of security is also emphasized when someone tries to do something accessing to the main information system. Accordingly, Biometrics has been used in many ways. OTP, however, must have a few devices accessing to several systems, and Biometrics involve some risk of mis-recognition rate and mis-denial rate. It also has the risk possible to access to the main information system when losing OTP. This research reduced risks about the loss as separating RFID leader for mobile, Tag and the accessor's cellular phone, and is about pseudo random validation key generated from the administration system through contact with RFID leader for mobile and Tag. As sending the key to user's cell phone which is already registered, security is strengthened more than existing connection methods through the ID and password. RFID for mobile not generalized to the present has been studied as a tool accessing to the main information system.

• Economic and Environmental Geology
• /
• v.41 no.6
• /
• pp.747-761
• /
• 2008

In order to experiment the acoustic remote classification of seabed sediment, we achieved ground-truth data(i.e. video and grab samples, etc.) and developed post-processing for automatic classification procedure on the basis of 300 kHz MultiBeam Echo Sounder(MBES) backscattering data, which was acquired using KONGBERG Simrad EM3000 at Sock-Cho Port, East Sea of South Korea. Sonar signal and its classification performance were identified with geo-referenced video imagery with the aid of GIS (Geographic Information System). The depth range of research site was from 5 m to 22.7 m, and the backscattering amplitude showed from -36dB to -15dB. The mean grain sizes of sediment from equi-distanced sampling site(50 m interval) varied from 2.86$(\phi)$ to 0.88(\phi). To acquire the main feature for the seabed classification from backscattering amplitude of MBES, we evaluated the correlation factors between the backscattering amplitude and properties of sediment samples. The performance of seabed remote classification proposed was evaluated with comparing the correlation of human expert segmentation to automatic algorithm results. The cross-model perception error ratio on automatic classification algorithm shows 8.95% at rocky bottoms, and 2.06% at the area representing low mean grain size.