• 정보처리학회논문지B
• /
• 제12B권1호
• /
• pp.9-16
• /
• 2005

본 논문은 얼굴인증 시스템 구현과 조명변화에 견인한 얼굴인증 방법들에 관한 연구에 초점을 둔다. 얼굴인증 시스템 구현을 위한 방법으로 PCA(Principal Component Analysis), GMM(Gaussian Mixture Models), 1차원 HMM(1 Dimensional Hidden Markov Models), 준 2차원 HMM(Pseudo 2 Dimensional Hidden Markov Models) 방법을 이용한다. 네 가지 다른 얼굴인증 방법들의 조명변화에 대한 성능비교 실험을 수행한다. 조명변화실험을 위해 얼굴이미지의 왼쪽에서 오른쪽으로 인공적인 조명효과(${\delta}=0,40,60,80$)를 준다. 얼굴특징벡터는 얼굴이미지에서 분할한 각 블록에 대한 2D DCT(2 Dimensional Discrete Cosine Transform) 계수를 이용하고 실험은 ORL(Olivetti Research Laboratory) 얼굴데이터베이스를 사용한다. 실험결과 모든 경우 조명변화 값이 커질수록 성능저하가 발생한다. 또한 조명변화가 없는 경우(${\delta}=0$) 준 2차원 HMM이 $2.54{\%}$, 1차원 HMM이 $3.18{\%}$, PCA가 $11.7{\%}$, GMM이 $13.38{\%}$의 EER(Equal Error Rate) 성능을 나타낸다. 조명변화가 없는 경우(${\delta}=0$) 1차원 HMM 방법이 PCA 방법보다 좋은 성능을 나타내지만 조명변화 ${\delta}{\geq}40$인 때에는 반대로 PCA 방법이 더 좋은 성능을 나타낸다. 마지막으로 준 2차원 HMM의 경우 조명변화에 관계없이 가장 좋은 EER성능을 나타낸다.

• 응용통계연구
• /
• 제27권6호
• /
• pp.1049-1068
• /
• 2014

자기회귀 모형(autoregressive model)은 일변량(univaraite) 시계열자료의 분석에서 널리 사용되는 방법 중 하나이다. 그러나 이 방법은 자료에 일정한 추세가 있다고 가정하기 때문에 자료에 분절(structural break)이 존재할 때 적절하지 않을 수 있다. 이러한 문제점을 해결하기 위한 방법으로 국면전환(regime-switching) 모형인 임계자기회귀 모형(threshold autoregressive model)이 제안되었는데 최근 지연 모수(delay parameter)을 포함한 이 국면전환(two regime-switching) 모형으로 확장되어 많은 연구가 활발히 진행되고 있다. 본 논문에서는 이 국면전환 임계자기회귀 모형을 베이지안(Bayesian) 관점에서 살펴본다. 베이지안 분석을 위해 모수적 임계자기 회귀 모형 뿐만 아니라 디리슐레 과정(Dirichlet Process) 사전분포를 이용하는 비모수적 임계자기 회귀 모형을 고려하도록 한다. 두 가지 베이지안 임계자기 회귀 모형을 바탕으로 사후분포를 유도하고 마코프 체인 몬테 카를로(Markov chain Monte Carlo) 방법을 통해 사후추론을 실시한다. 모형 간의 성능을 비교하기 위해 모의실험을 통한 자료 분석을 고려하고, 더 나아가 한국과 미국의 국내 총생산(Gross Domestic Product)에 대한 실증적 자료 분석을 실시한다.

• 대한음성학회지:말소리
• /
• 제62호
• /
• pp.113-131
• /
• 2007

It is reported that the segmental feature based recognition system shows better results than conventional feature based system in the previous studies. On the other hand, the various studies of combining neural network and hidden Markov models within a single system are done with expectations that it may potentially combine the advantages of both systems. With the influence of these studies, tandem approach was presented to use neural network as the classifier and hidden Markov models as the decoder. In this paper, we applied the trend information of segmental features to tandem architecture and used posterior probabilities, which are the output of neural network, as inputs of recognition system. The experiments are performed on Auroral database to examine the potentiality of the trend feature based tandem architecture. From the results, the proposed system outperforms on very low SNR environments. Consequently, we argue that the trend information on tandem architecture can be additionally used for traditional MFCC features.

• 한국음향학회:학술대회논문집
• /
• 한국음향학회 1994년도 제11회 음성통신 및 신호처리 워크샵 논문집 (SCAS 11권 1호)
• /
• pp.395-398
• /
• 1994

In this paper the word recognition using sub-model based Hidden Markov Model was studied. Phoneme models were composed of 61 phonemes in therms of Korean language pronunciation characteristic. Using this, word model was maded by serial concatenation. But, in case of this phoneme concatenation, the second and the third phoneme of syllable are overlapped in distribution at the same time. So considering this, the method that combines the second and the third phoneme to one model was proposed. And to prevent the increase in number of model, similar phonemes were combined to one, and finially, 57 models were created. In experiment proper model structure of sub-model was searched for, and recognition results were compared. So similar recognition results were maded, and overall recognition rates were increased in case of using parameter tying method.

• Journal of the Korean Data and Information Science Society
• /
• 제27권6호
• /
• pp.1621-1629
• /
• 2016

This paper studies Bayesian pooling for analysis of categorical data from small areas. Many surveys consist of categorical data collected on a contingency table in each area. Statistical inference for small areas requires considerable care because the subpopulation sample sizes are usually very small. Typically we use the hierarchical Bayesian model for pooling subpopulation data. However, the customary hierarchical Bayesian models may specify more exchangeability than warranted. We, therefore, investigate the effects of pooling in hierarchical Bayesian modeling for the contingency table from small areas. In specific, this paper focuses on the methods of direct or indirect pooling of categorical data collected on a contingency table in each area through Dirichlet priors. We compare the pooling effects of hierarchical Bayesian models by fitting the simulated data. The analysis is carried out using Markov chain Monte Carlo methods.

• 대한산업공학회지
• /
• 제34권2호
• /
• pp.140-150
• /
• 2008

Point Coordination Function (PCF) of the IEEE 802.11 protocol providing a centrally-controlled polling-based multiple access to a wireless channel is very efficient in high load conditions. However, its performance degrades with increasing the number of terminals and decreasing the load, because of wastes related to unsuccessful polling attempts. To solve the problem, we propose and study analytically the generic dynamic polling policy using backoff concept. For this aim, we develop Markov models describing the network queues changes, what allows us to estimate such performance measures as the average MAC service time and the average MAC sojourn time, to show the dynamic polling efficiency and to tune optimally the backoff rule.

• 한국경영과학회지
• /
• 제23권4호
• /
• pp.97-109
• /
• 1998

In this paper, we propose a new approach to solve stochastic fluid flow models applied to the analysis of ceil loss of an ATM multiplexer. Existing stochastic fluid flow models have been analyzed by using linear differential equations. In case of large state space, however. analyzing stochastic fluid flow model without numerical errors is not easy. To avoid this numerical errors and to analyze stochastic fluid flow model with large state space. we develope a new computational algorithm. Instead of solving differential equations directly, this approach uses iterative and numerical method without calculating eigenvalues. eigenvectors and boundary coefficients. As a result, approximate solutions and upper and lower bounds are obtained. This approach can be applied to stochastic fluid flow model having general Markov chain structure as well as to the superposition of heterogeneous ON-OFF sources it can be extended to Markov process having non-exponential sojourn times.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2007년도 한국음성과학회 공동학술대회 발표논문집
• /
• pp.323-326
• /
• 2007

The various studies of combining neural network and hidden Markov models within a single system are done with expectations that it may potentially combine the advantages of both systems. With the influence of these studies, tandem approach was presented to use neural network as the classifier and hidden Markov models as the decoder. In this paper, we applied the trend information of segmental features to tandem architecture and used posterior probabilities, which are the output of neural network, as inputs of recognition system. The experiments are performed on Aurora2 database to examine the potentiality of the trend feature based tandem architecture. The proposed method shows the better results than the baseline system on very low SNR environments.

• 전자공학회논문지B
• /
• 제32B권5호
• /
• pp.756-764
• /
• 1995

In this paper, we proposed MLP(MultiLayer Perceptron) based HMM's(Hidden Markov Models) with bounded state duration for isolated word recognition. The minimum and maximum state duration for each state of a HMM are estimated during the training phase and used as parameters of constraining state transition in a recognition phase. The procedure for estimating these parameters and the recognition algorithm using the proposed HMM's are also described. Speaker independent isolated word recognition experiments using a vocabulary of 10 city names and 11 digits indicate that recognition rate can be improved by adjusting the minimum state durations.

• 대한수학회논문집
• /
• 제14권2호
• /
• pp.405-423
• /
• 1999

We consider a statistical multiplexer model with finite buffer capacity and finite number of independent identical 3-state bursty voice sources. The burstiness of the sources is modeled by describing both two different active periods (at the rate of one packet perslot) and the passive periods during which no packets are generated. Assuming a mixture of two geometric distributions for active period and a geometric distribution for passive period and geometric distribution for passive period, we derive the recursive algorithm for the probability mass function of the buffer contents (in packets). We also obtain loss probability and the distribution of packet delay. Numerical results show that the system performance deteriorates considerably as the variance of the active period increases. Also, we see that the loss probability of 2-state Markov models is less than that of 3-state Markov models.