• 한국컴퓨터정보학회논문지
• /
• 제19권1호
• /
• pp.23-30
• /
• 2014

이미지 기술자(descriptor)를 이용한 정합은 최근까지 컴퓨터 비전과 패턴인식 분야에서 사용되고 있는 강력한 정합 방법이다. 그러나 3차원 시점이 변화되거나 밝기가 변화된 이미지, 반복된 패턴이 포함된 이미지 등에서 잘못된 정합들이 발생한다. 본 논문에서는 반복된 패턴이 포함되어 있는 이미지에서 잘못된 정합들이 많이 발생하는 문제점에 대해 기술하고 이를 분석하여 잘못된 정합들을 제거할 수 있는 방법을 제안한다. MDMF(Multiple Descriptors-based Mismatch Filtering) 방법은 각 특징점에 대해 인접한 여러 개의 특징점들의 기술자들을 사용하여 다중 기술자를 생성한 후 이를 활용하여 잘못된 정합들을 제거한다. 실험에서는 크기 변환, 회전 변환, 어파인 변환에 대해 기존 SIFT와 ASIFT의 정합율을 MDMF를 이용해 제거한 정합율과 비교하여 MDMF가 잘못된 정합을 성공적으로 제거할 수 있음을 보였다.

• ETRI Journal
• /
• 제30권5호
• /
• pp.753-755
• /
• 2008

In this letter, we propose a new histogram equalization technique for feature compensation in speech recognition under noisy environments. The proposed approach combines a signal-to-noise-ratio-dependent feature reconstruction method and the class histogram equalization technique to effectively reduce the acoustic mismatch present in noisy speech features. Experimental results from the Aurora 2 task confirm the superiority of the proposed approach for acoustic feature compensation.

• 말소리와 음성과학
• /
• 제12권3호
• /
• pp.65-71
• /
• 2020

특징 정규화는 음성 특징 파라미터들의 통계적인 특성의 정규화를 통해 훈련 및 테스트 조건 사이의 환경 불일치의 영향을 감소시키는 방법으로서 기존의 Gaussian mixture model-hidden Markov model(GMM-HMM) 기반의 음성인식 시스템에서 우수한 성능개선을 입증한 바 있다. 하지만 심층신경망(deep neural network, DNN) 기반의 음성인식 시스템에서는 환경 불일치의 영향을 최소화 하는 것이 반드시 최고의 성능 개선으로 연결되지는 않는다. 본 논문에서는 이러한 현상의 원인을 과도한 특징 정규화로 인한 정보손실 때문이라 보고, 음향모델을 훈련 하는데 유용한 정보는 보존하면서 환경 불일치의 영향은 적절히 감소시켜 음성인식 성능을 최대화 하는 특징 정규화 방식이 있는 지 검토해보고자 한다. 이를 위해 평균 정규화(mean normalization, MN)와 평균 및 분산 정규화(mean and variance normalization, MVN)의 절충 방식인 평균 및 지수적 분산 정규화(mean and exponentiated variance normalization, MEVN)를 도입하여, 잡음 및 잔향 환경에서 분산에 대한 정규화의 정도에 따른 DNN 기반의 음성인식 시스템의 성능을 비교한다. 실험 결과, 성능 개선의 폭이 크지는 않으나 분산 정규화의 정도에 따라 MEVN이 MN과 MVN보다 성능이 우수함을 보여준다.

• 한국멀티미디어학회논문지
• /
• 제19권5호
• /
• pp.848-855
• /
• 2016

Face recognition has received a great deal of attention for the wide range of applications in real-world scenario. In this scenario, mismatches (so called heterogeneity) in terms of resolution and illumination between gallery and test face images are inevitable due to the different capturing conditions. In order to deal with the mismatch problem, we propose a local feature learning method using deep canonical correlation analysis (DCCA) for heterogeneous face recognition. By the DCCA, we can effectively reduce the mismatch between the gallery and the test face images. Furthermore, the proposed local feature learned by the DCCA is able to enhance the discriminative power by using facial local structure information. Through the experiments on two different scenarios (i.e., matching near-infrared to visible face images and matching low-resolution to high-resolution face images), we could validate the effectiveness of the proposed method in terms of recognition accuracy using publicly available databases.

• The Journal of the Acoustical Society of Korea
• /
• 제24권1E호
• /
• pp.21-27
• /
• 2005

Environmental mismatch has been one of important issues discussed in matched field processing for underwater source detection problem. To overcome this mismatch many algorithms professing robustness have been suggested. Feature extraction method (FEM) [Seong and Byun, IEEE Journal of Oceanic Engineering, 27(3), 642-652 (2002)] is one of robust matched field processing algorithms, which is based on the eigenvector estimation. Excluding eigenvectors of replica covariance matrix corresponding to large eigenvalues and forming an incoherent subspace of the replica field, the processor is formulated similarly to MUSIC algorithm. In this paper, by using the ocean experimental data, processing results of FEM and MVDR with white noise constraint (WNC) are presented for two levels of multi-tone source. Analysis of eigen-space of CSDM and FEM performance are also presented.

• 한국언어정보학회:학술대회논문집
• /
• 한국언어정보학회 2007년도 정기학술대회
• /
• pp.234-240
• /
• 2007

This paper argues that various kinds of displaced structures in English should be licensed by a more explicitly formulated type of rule schema in order to deal with what is called weak connectivity in English. This paper claims that the filler and the gap site cannot maintain the total identity of features but a partial overlap since the two positions need to obey the structural forces that come from occupying respective positions. One such case is the missing object construction where the subject fillers and the object gaps are to observe requirements that are imposed on the respective positions. Others include passive constructions and topicalized structures. In this paper, it is argued that the feature discrepancy comes from the different syntactic positions in which the fillers are assumed to be located before and after displacement. In order to capture this type of mismatch, syntactically relevant features are handled separately from the semantically motivated features in order to deal with the syntactically imposed requirements.

• 한국정보통신학회논문지
• /
• 제14권3호
• /
• pp.613-620
• /
• 2010

화자인증 시스템은 훈련 환경과 인식 환경이 다른 경우 인식 성능이 크게 저하된다. 이러한 훈련과 인식 환경의 불일치는 다양한 잡음과 상이한 채널 환경 때문이다. 본 논문은 화자인증 시스템의 강인성 개선을 위하여 음성신호의 위상에 기반한 특정 추출 기법을 제안한다. 이 방법은 음성신호의 위상으로부터 순시 주파수를 계산하여 대역별로 순시 주파수를 모두 모아 구한 히스토그램으로부터 특징 계수를 추출한다. 이 특징 파라미터를 적용한 결과 조 용한 환경뿐만 아니라 잡음환경 그리고 채널 왜곡 환경에서도 화자인증 시스템의 성능이 개선됨을 알 수 있다.

• 음성과학
• /
• 제11권1호
• /
• pp.7-20
• /
• 2004

In this paper, we propose a two-step noise compensation algorithm in feature extraction for achieving robust speech recognition. The proposed method frees us from requiring a priori information on noisy environments and is simple to implement. First, in frequency domain, the Harmonics-based Spectral Subtraction (HSS) is applied so that it reduces the additive background noise and makes the shape of harmonics in speech spectrum more pronounced. We then apply a judiciously weighted variance Feature Vector Normalization (FVN) to compensate for both the channel distortion and additive noise. The weighted variance FVN compensates for the variance mismatch in both the speech and the non-speech regions respectively. Representative performance evaluation using Aurora 2 database shows that the proposed method yields 27.18% relative improvement in accuracy under a multi-noise training task and 57.94% relative improvement under a clean training task.

• 대한음성학회지:말소리
• /
• 제53호
• /
• pp.103-118
• /
• 2005

The performance of speech recognition system is degraded by mismatch between training and test environments. Recently, Stereo-based Piecewise LInear Compensation for Environments (SPLICE) was introduced to overcome environmental mismatch using stereo data. In this paper, we propose several methods to improve the conventional SPLICE and evaluate them in the Aurora2 task. We generalize SPLICE to compensate for covariance matrix as well as mean vector in the feature space, and thereby yielding the error rate reduction of 48.93%. We also employ the weighted sum of correction vectors using posterior probabilities of all Gaussians, and the error rate reduction of 48.62% is achieved. With the combination of the above two methods, the error rate is reduced by 49.61% from the Aurora2 baseline system.

• 센서학회지
• /
• 제27권6호
• /
• pp.374-379
• /
• 2018

The piezocomposite transducer is widely used because it is highly efficient in transforming electric energy into mechanical energy, and its frequency range is broader than that of other types of ultrasound transducers. A general piezocomposite transducer is composed of an acoustic lens, impedance matching layers, piezoelectric materials, and backing layers. When an input voltage is applied to a piezoelectric material as an active material, it generates sound waves while vibrating. At that time, an impedance matching layer helps the sound waves to propagate forward while reducing the impedance mismatch that may occur at the interface between the active material and its front material. The impedance mismatch has a negative effect on the signal of an ultrasound transducer; thus, it is important to design a matching layer to overcome the issue. In this study, an optimized feature of a matching layer with gradient properties is studied. An objective function is defined to minimize both the average and the deviation of the reflection coefficients that are functions of the frequencies. As a result, an improvement in the signal characteristics with respect to the sensitivity and bandwidth is reported.