• Proceedings of the Korea Information Processing Society Conference
• /
• 2010.11a
• /
• pp.762-765
• /
• 2010

특정 시간에 동시 연주된 다수 음원의 onset 을 검출하기 위해서는 음원 분리 문제가 선결되어야 한다. 특히, 드럼과 같은 조음(?音) 악기 신호 검출 문제를 해결하기 위해서는 음원 분리 방법의 성능이 중요하다. 이에 본 연구에서는 효과적인 음원 분리 방법으로 알려진 확률적 은닉 성분 분석(PLCA) 방법에 기반한 주요 악기 신호의 onset 검출 방법을 제안한다. 효과적인 onset 검출을 위해, 첫째, 확률적 은닉 성분 분석으로 훈련 된 비음수 주파수 성분 중 최적의 성분을 선택하는 방법을 적용하고, 둘째, 드럼 악기 신호의 정확한 onset 검출을 위해 고안된 비음수 시계열 신호 threshold 방법을 적용한다. 실험에서는 제시된 방법을 이용하여 드럼의 주요 악기 신호 onset 검출 성능이 향상됨을 보인다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.37 no.2C
• /
• pp.181-187
• /
• 2012

Millimeter wave (MMW) readily penetrates fabrics, thus it can be used to detect objects concealed under clothing. A passive MMW imaging system can operate as a stand-off type sensor that scans people in both indoors and outdoors. However, because of the diffraction limit and low signal level, the imaging system often suffers from low image quality. Therefore, suitable statistical analysis and computational processing would be required for automatic analysis of the images. In this paper, a real-time concealed object detection is addressed by means of the multi-level segmentation. The histogram of the image is modeled with a Gaussian mixture distribution, and hidden object areas are segmented by a multi-level scheme involving $k$-means, the expectation-maximization algorithm, and a decision rule. The complete algorithm has been implemented in C++ environments on a standard computer for a real-time process. Experimental and simulation results confirm that the implemented system can achieve the real-time detection of concealed objects.

• The KIPS Transactions:PartB
• /
• v.11B no.2
• /
• pp.199-206
• /
• 2004

This paper proposes a hybrid independent component analysis(ICA) of fixed-point(FP) algorithm and robust algorithm. The FP algorithm is applied for improving the analysis speed and performance, and the robust algorithm is applied for preventing performance degradations by means of very small kurtosis and temporal correlations between components. And the adaptive adaptation of temporal correlations has been proposed for solving limits of the conventional robust algorithm dependent on the maximum time delay. The proposed ICA has been applied to the problems for separating the 4-mixed signals of 500 samples and 10-mixed images of $512\times512$pixels, respectively. The experimental results show that the proposed ICA has a characteristics of adaptively adapting the maximum time delay, and has a superior separation performances(speed, rate) to conventional FP-ICA and hybrid ICA of heuristic correlation. Especially, the proposed ICA gives the larger degree of improvement as the problem size increases.

• The KIPS Transactions:PartB
• /
• v.11B no.3
• /
• pp.381-386
• /
• 2004

This paper proposes an independent component analysis(ICA) of the fixed-point(FP) algorithm based on Newton method by adding the kurtosis. The kurtosis is applied for clustering the components, and the FP algorithm of Newton method is applied for improving the analysis speed and performance. The proposed ICA has been applied to the problems for separating the 6-mixed signals of 500 samples and 8-mixed images of $512\times512$pixels, respectively. The experimental results show that the proposed ICA has always a fixed analysis sequence. The result can be solved the limit of conventional ICA which has a variable sequence depending on the running of algorithm. Especially, the proposed ICA can be used to classify and identify the signals or the images.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1998.06e
• /
• pp.327-330
• /
• 1998

독립성분분석(ICA: Independent Component Analysis)이란 특징이 상이한 둘 이상의 신호들이 선형적으로 결합되어 있을 때 이를 효과적으로 분리하는 방법들을 통칭하며 잡음제거, 음질개선 및 신호처리 분야에서 많이 활용되고 있다. 본 논문에서는 전화음성 화자인식 시스템의 성능향상을 위해 독립성분분석을 이용하는 방법을 제안한다. 먼저 화자가 발성한 음성신호의 켑스트럼 계수를 여러 채널 함수들의 선형적인 합으로 가정하고, 독립성분분석을 이용하여 얻은 새로운 켑스트럼 벡터를 학습과 인식에 사용하였다. 실험자료는 잔화음성 화자식별기의 성능평가에 널리 쓰이고 있는 SPIDRE를 사용하였고 regodic 은닉 마코프 모델을 이용하여 문장 독립 화자식별 시스템을 구성하였다. 학습음성의 특징과 실험음성의 특징이 다른 조건에서 기존의 채널 정규화 방법들에 비해 10~15%이상 인식률이 향상되었다.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1998.06e
• /
• pp.399-402
• /
• 1998

본 논문에서는 환자의 음성을 정상, 양성종양, 악성종양으로 분류하는 실험을 켑스트럼 파라미터를 통한 음원분리와 신경회로망을 이용하여 수행하고 그 결과를 보고한다. 기존의 장애음성 데이터베이스에는 정상음성과 양성종양의 경우만 수록되어 있었고 외국의 환자들을 대상으로 한 경우만 있었기 때문에 국내의 환자들에게 직접 적용할 경우 어떠한 결과가 나올지 예측하기가 어려웠다. 최근 부산대학교 이비인후과팀에서 수집한 국내의 정상, 양성, 악성종양의 경우에 대한 데이터베이스를 분석하고 신경회로망에 의해 분류함으로써 사람의 음성신호만에 의한 후두질환이 식별이 가능하였다. 본 실험에서는 식별 파라미터로 음성신호의 선형예측오차신호에 관한 켑스트럼으로부터 음원비인 HNRR을 구하여 Jitter, Shimmer와 함께 사용하였다. 신경회로망은 입, 출력 층과 한 개의 은닉층을 갖는 다층신경망을 이용하였으며, 식별은 두단계로 나누어 정상과 비정상을 분류한 후 다시 비정상을 양성과 악성으로 분류하였다[1].

• The Journal of the Acoustical Society of Korea
• /
• v.38 no.4
• /
• pp.427-432
• /
• 2019

In this paper, we propose a home monitoring system using sound event detection based on a bidirectional gated recurrent neural network for the hard-of-hearing. First, in the proposed system, packet loss concealment is used to recover a lost signal captured through wireless sensor networks, and reliable channels are selected using multi-channel cross correlation coefficient for effective sound event detection. The detected sound event is converted into the text and haptic signal through a harmonic/percussive sound source separation method to be provided to hearing impaired people. Experimental results show that the performance of the proposed sound event detection method is superior to the conventional methods and the sound can be expressed into detailed haptic signal using the source separation.

• Journal of the Korean Society of Industry Convergence
• /
• v.7 no.1
• /
• pp.41-46
• /
• 2004

This paper proposes a separation of mixed signals by using the robust independent component analysis(RICA) based on neural networks. RICA is based on the temporal correlations and the second order statistics of signal. This method e is applied for improving the analysis rate and speed in which the sources have very small or zero kurtosis. The proposed method has been applied for separating the 10 mixed finger prints of $256{\times}256$-pixel and the 4 mixed images of $512{\times}512$-pixel, respectively. The simulation results show that RICA has the separating rate and speed better than those using the conventional FP algorithm based on Newton method.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.36C no.12
• /
• pp.78-89
• /
• 1999

Previous wavelet network structures are determined by considering the relationship between wavelet windows distribution of training patterns that are transformed into time-frequency space. Because it is separated two algorithms that determines wavelet network structure and that modifies parameters of network, learning process that minimizes output error of network is executed after the network structure is determined. But this method has some weakness that training patterns must be transformed into time-frequency space by additional preprocessing and the network structure should be fixed during learning process. In this paper, we propose a new constructing method for wavelet network structure by using differences between the output and the desired response without preprocessing. Because the algorithm perform network construction and error minimizing process simultaneously, it can determine the number of hidden nodes adaptively as with the complexity of problems. In addition, the network structure is optimized by inserting new hidden nodes in the area that has maximum error and extracting hidden nodes that has no effect to the output of network. This algorithm has no constraint condition that all training patterns must be known, because it removes preprocessing procedure for training patterns and it can be applied effectively to systems that has time varying outputs.

• Smart Media Journal
• /
• v.1 no.2
• /
• pp.12-22
• /
• 2012

Today we are witnessing an increasingly widespread use of cameras in our lives for video surveillance, robot vision, and mobile phones. This has led to a renewed interest in computer vision in general and an on-going boom in human activity recognition in particular. Although not particularly fancy per se, human gait is inarguably the most common and frequent action. Early on this decade there has been a passing interest in human gait recognition, but it soon declined before we came up with a systematic analysis and understanding of walking motion. This paper presents a set of DBN-based models for the analysis of human gait in sequence of increasing complexity and modeling power. The discussion centers around HMM-based statistical methods capable of modeling the variability and incompleteness of input video signals. Finally a novel idea of extending the discrete state Markov chain with a continuous density function is proposed in order to better characterize the gait direction. The proposed modeling framework allows us to recognize pedestrian up to 91.67% and to elegantly decode out two independent gait components of direction and posture through a sequence of experiments.