• 한국지능시스템학회논문지
• /
• 제21권4호
• /
• pp.512-517
• /
• 2011

얼굴영상에서 나타나는 정서특징을 분석하기 위하여 본 논문에서는 Active Shape Model (ASM)과 Lucas-Kanade (LK) optical flow 기법을 기반으로 하는 특징검출 및 분석방법을 제안한다. Facial Action Coding System에 근거하여 묘사된 정서적 특징을 고려하여, 특징이 분포하는 영역에 위치한 다수의 landmark로 shape 모델을 구성하고 모델에서 각 Landmark를 중심으로 하는 움직임 벡터 윈도우 내부의 픽셀에 대한 LK 기법을 통해 optical flow 벡터를 추출한다. 추출된 움직임 벡터의 방향성 조합에 근거하여 얼굴정서특징을 shape 모델로 표현할 수 있으며, 베이지안 분류기라는 확률 기반 추론기법을 기반으로 정서적 상태에 대한 추정할 수 있다. 또한, 정서특징분석과정의 연산 효율성과 정확성 향상을 도모하기 위하여 common spatial pattern (CSP) 분석기법을 적용하여 정서상태 별로 상관성이 높은 특징만으로 구성된 최적정서특징을 추출한다.

• 방송공학회논문지
• /
• 제24권6호
• /
• pp.1113-1121
• /
• 2019

본 논문에서는 얼굴의 밝기와 색상 정보를 함께 이용한 합성곱 신경망 기반의 얼굴 위변조 검출 방법을 제안한다. 제안하는 방법은 적층된 합성곱 신경망과 보조 신경망을 이용하여 실제 얼굴과 위변조된 얼굴의 밝기 특징과 색상 특징을 독립적으로 추출한다. 기존의 방법과는 달리, 본 논문에서는 추출된 특징을 단순 결합(Concatenation)하는 것이 아니라 주의 모듈(Attention Module)을 이용하여 적응적(Adaptively)으로 조합할 수 있도록 하였다. 또한, 효과적인 분류기 학습을 위하여 대비 손실함수(Contrast Loss Function)를 새롭게 제안하였는데, 대비 손실함수는 동일 클래스 내의 특징 간의 차이는 최소화 시키고 서로 다른 클래스의 특징 간의 차이는 최대화 시킴으로써 특징의 분별력을 높인다. 다양한 실험을 통해 본 논문에서 제안하는 방법이 기존 얼굴 위변조 검출 방법 대비 개선된 성능을 보임을 확인하고 그 결과를 분석한다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제7권1호
• /
• pp.99-118
• /
• 2013

Electrocardiograms (ECGs) are widely used by clinicians to identify the functional status of the heart. Thus, there is considerable interest in automated systems for real-time monitoring of arrhythmia. However, intra- and inter-patient variability as well as the computational limits of real-time monitoring poses significant challenges for practical implementations. The former requires that the classification model be adjusted continuously, and the latter requires a reduction in the number and types of ECG features, and thus, the computational burden, necessary to classify different arrhythmias. We propose the use of adaptive learning to automatically train the classifier on up-to-date ECG data, and employ adaptive feature selection to define unique feature subsets pertinent to different types of arrhythmia. Experimental results show that this hybrid technique outperforms conventional approaches and is therefore a promising new intelligent diagnostic tool.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2005년도 ICCAS
• /
• pp.1431-1434
• /
• 2005

In the steel production line, the molten metal of a furnace is transformed into billet and then moves to the heating furnace of the hot rolling mill. This paper describes about the development of recognition system for the characters, which was marked at the billet material by use template-marking plate and hand written method, in the steel plant. For the recognition of template-marked characters, we propose PSVM algorithm. And for the recognition of hand written character, we propose combination methods of CCD algorithm and PSVM algorithm. The PSVM algorithm need some more time than the conventional KLT or SVM algorithm. The CCD algorithm makes shorter classification time than the PSVM algorithm and good for the classification of closed curve characters from Arabic numerals. For the confirmation of algorithm, we have compared our algorithm with conventional methods such as KLT classifier and one-to-one SVM. The recognition rate of experimented billet characters shows that the proposing PSVM algorithm is 97 % for the template-marked characters and combinational algorithm of CCD & PSVM is 95.5 % for the hand written characters. The experimental results show that our proposing method has higher recognition rate than that of the conventional methods for the template-marked characters and hand written characters. By using our algorithm, we have installed real time character recognition system at the billet processing line of the steel-iron plant.

• 대한전자공학회논문지SP
• /
• 제41권6호
• /
• pp.175-186
• /
• 2004

본 논문은 복잡한 컬러 영상에서의 문자 추출을 위한 텍스춰와 연결성분 방법의 결합된 방법을 제안한다. 자동 학습 방법으로 구축된 다층 신경망(multilayer perceptron)은 부트스트랩 학습 방법을 사용함으로써 별도의 특징값 추출 단계 없이 다양한 환경의 입력 영상에 대한 검출률(recall rate)을 향상시키며, 검출률을 향상함으로써 발생되는 정확도(precision rate) 저하 문제는, NMF(Non-negative matrix factorization)를 이용한 연결 성분 방법을 사용함으로써 극복한다. 문자의 존재 비율이 낮은 입력영상에 대하여 CAMShift 알고리즘을 이용한 영역 마킹 방법을 사용함으로써, 두 방법을 결합함으로써 야기되는 속도 저하 문제의 해결을 시도하였다. 이와 같이 텍스춰와 연결성분 방법을 결합함으로써 강건하고 효율적인 시스템을 구성할 수 있었다.

• ETRI Journal
• /
• 제39권2호
• /
• pp.191-201
• /
• 2017

This paper proposes an approach using taxonomic relatedness for answer-type recognition and type coercion in a question-answering system. We introduce a question analysis method for a lexical answer type (LAT) and semantic answer type (SAT) and describe the construction of a taxonomy linking them. We also analyze the effectiveness of type coercion based on the taxonomic relatedness of both ATs. Compared with the rule-based approach of IBM's Watson, our LAT detector, which combines rule-based and machine-learning approaches, achieves an 11.04% recall improvement without a sharp decline in precision. Our SAT classifier with a relatedness-based validation method achieves a precision of 73.55%. For type coercion using the taxonomic relatedness between both ATs and answer candidates, we construct an answer-type taxonomy that has a semantic relationship between the two ATs. In this paper, we introduce how to link heterogeneous lexical knowledge bases. We propose three strategies for type coercion based on the relatedness between the two ATs and answer candidates in this taxonomy. Finally, we demonstrate that this combination of individual type coercion creates a synergistic effect.

• 전기학회논문지
• /
• 제66권4호
• /
• pp.692-700
• /
• 2017

In this paper, we propose the structural design of Interval Type-2 FCM based RBFNN. Proposed model consists of three modules such as condition, conclusion and inference parts. In the condition part, Interval Type-2 FCM clustering which is extended from FCM clustering is used. In the conclusion part, the parameter coefficients of the consequence part are estimated through LSE(Least Square Estimation) and WLSE(Weighted Least Square Estimation). In the inference part, final model outputs are acquired by fuzzy inference method from linear combination of both polynomial and activation level obtained through Interval Type-2 FCM and acquired activation level through Interval Type-2 FCM. Additionally, The several parameters for the proposed model are identified by using differential evolution. Final model outputs obtained through benchmark data are shown and also compared with other already studied models' performance. The proposed algorithm is performed by using Iris and Vehicle data for pattern classification. For the validation of regression problem modeling performance, modeling experiments are carried out by using MPG and Boston Housing data.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2007년도 Proceedings of ISRS 2007
• /
• pp.6-10
• /
• 2007

Classification of hyperspectral images is challenging. A very high dimensional input space requires an exponentially large amount of data to adequately and reliably represent the classes in that space. In other words in order to obtain statistically reliable classification results, the number of necessary training samples increases exponentially as the number of spectral bands increases. However, in many situations, acquisition of the large number of training samples for these high-dimensional datasets may not be so easy. This problem can be overcome by using multiple classifiers. In this paper we compared the effectiveness of two approaches for creating multiple classifiers, feature selection and feature extraction. The methods are based on generating multiple feature subsets by running feature selection or feature extraction algorithm several times, each time for discrimination of one of the classes from the rest. A maximum likelihood classifier is applied on each of the obtained feature subsets and finally a combination scheme was used to combine the outputs of individual classifiers. Experimental results show the effectiveness of feature extraction algorithm for generating multiple classifiers.

• 한국콘텐츠학회논문지
• /
• 제5권6호
• /
• pp.41-48
• /
• 2005

이 논문은 얼굴인식을 수행하기 위해서 이미 잘 알려진 주성분 분석법과 선형판별 분석법에 레이디얼 기저 함수 신경망을 결합한 인식 알고리즘을 제시하였다. 입력된 원래의 얼굴영상은 주성분분석법을 통하여 차원을 줄인 고유 얼굴 가중치를 산출한다. 이 가중치 벡터를 선형판별 분석법의 입력데이터로 사용하여 선형판별분석의 변환행렬을 계산할 때 클래스 내의 분산행렬에서 특이점이 발생하지 않도록 하면서 특징벡터를 산출하여 인식을 수행하였다. 두 번째 시도에서는 선형판별분석법에 의해 생성된 특징벡터를 레이디얼 기저 함수 신경망에 입력하여 학습하고 얼굴인식을 수행하였다. ORL DB의 얼굴영상에 대해 실험한 결과 93.5%의 인식률을 얻을 수 있었다.

• 말소리와 음성과학
• /
• 제6권2호
• /
• pp.21-28
• /
• 2014

We propose a novel scheme for digital audio authentication of given audio files which are edited by inserting small audio segments from different environmental sources. The purpose of this research is to detect inserted sections from given audio files. We expect that the proposed method will assist human investigators by notifying suspected audio section which considered to be recorded or transmitted on different environments. GMM-UBM and GSV-SVM are applied for modeling the dominant environment of a given audio file. Four kinds of likelihood ratio based scores and SVM score are used to measure the likelihood for a dominant environment model. We also use an ensemble score which is a combination of the aforementioned five kinds of scores. In the experimental results, the proposed method shows the lowest average equal error rate when we use the ensemble score. Even when dominant environments were unknown, the proposed method gives a similar accuracy.