• Journal of Advanced Navigation Technology
• /
• v.13 no.1
• /
• pp.151-155
• /
• 2009

We proposes a new method using the non-negative matrix factorization (NMF) and clustering method to extract the sentences for personalized document summarization. The proposed method uses clustering method for retrieving documents to extract sentences which are well reflected topics and sub-topics in document. Beside it can extract sentences with respect to query which are well reflected user interesting by using the inherent semantic features in document by NMF. The experimental results shows that the proposed method achieves better performance than other methods use the similarity and the NMF.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.3
• /
• pp.55-62
• /
• 2017

Non-negative matrix factorization (NMF) is one of the typical parts-based representation in which images are expressed as a linear combination of basis vectors that show the lcoal features or objects in the images. In this paper, we represent face images using various NMF methods and recognize their face identities based on extracted features using a learning vector quantization. We analyzed the various NMF methods by comparing extracted basis vectors. Also we confirmed the availability of NMF to the face recognition by verification of recognition rate of the various NMF methods.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.4
• /
• pp.468-479
• /
• 2022

To detect targets with active sonar system in the underwater environments, the targets are localized by receiving the echoes of the transmitted sounds reflected from the targets. In this case, reverberation from the scatterers is also generated, which prevents detection of the target echo. To detect the target effectively, reverberation suppression techniques such as pre-whitening based on autoregressive model and principal component inversion have been studied, and recently a Non-negative Matrix Factorization (NMF)-based technique has been also devised. The NMF-based reverberation suppression technique shows improved performance compared to the conventional methods, but the geometry of the transducer and receiver and attenuation by distance have not been considered. In this paper, the performance is improved through preprocessing such as the directionality of the receiver, Doppler related thereto, and attenuation for distance, in the case of using a continuous wave with a bistatic sonar. In order to evaluate the performance of the proposed system, simulation with a reverberation model was performed. The results show that the detection probability performance improved by 10 % to 40 % at a low false alarm probability of 1 % relative to the conventional non-negative matrix factorization.

• Journal of KIISE:Software and Applications
• /
• v.31 no.10
• /
• pp.1412-1417
• /
• 2004

Sparse coding or independent component analysis (ICA) which is a holistic representation, was successfully applied to elucidate early auditor${\gamma}$ processing and to the task of sound classification. In contrast, parts-based representation is an alternative way o) understanding object recognition in brain. In this thesis we employ the non-negative matrix factorization (NMF) which learns parts-based representation in the task of sound classification. Methods of feature extraction from the spectro-temporal sounds using the NMF in the absence or presence of noise, are explained. Experimental results show that NMF-based features improve the performance of sound classification over ICA-based features.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.2
• /
• pp.120-130
• /
• 2020

This paper solves the problem in underdetermined convolutive mixture by improving the disadvantages of the multichannel nonnegative matrix factorization technique widely used in blind source separation. In conventional researches based on Spatial Covariance Matrix (SCM), each element composed of values such as power gain of single channel and correlation tends to degrade the quality of the separated sources due to high variance. In this paper, level and frequency normalization is performed to effectively cluster the estimated sources. Therefore, we propose a novel SCM and an effective distance function for cluster pairs. In this paper, the proposed SCM is used for the initialization of the spatial model and used for hierarchical agglomerative clustering in the bottom-up approach. The proposed algorithm was experimented using the 'Signal Separation Evaluation Campaign 2008 development dataset'. As a result, the improvement in most of the performance indicators was confirmed by utilizing the 'Blind Source Separation Eval toolbox', an objective source separation quality verification tool, and especially the performance superiority of the typical SDR of 1 dB to 3.5 dB was verified.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.8
• /
• pp.39-48
• /
• 2023

This study was conducted to provide a solution to the problem of sizing errors occurring in online purchases due to discrepancies and non-standardization in clothing sizes. This paper discusses an implementation approach for a machine learning-based recommender system capable of providing personalized sizes to online consumers. We trained multiple validated collaborative filtering algorithms including Non-Negative Matrix Factorization (NMF), Singular Value Decomposition (SVD), k-Nearest Neighbors (KNN), and Co-Clustering using purchasing data derived from online commerce and compared their performance. As a result of the study, we were able to confirm that the NMF algorithm showed superior performance compared to other algorithms. Despite the characteristic of purchase data that includes multiple buyers using the same account, the proposed model demonstrated sufficient accuracy. The findings of this study are expected to contribute to reducing the return rate due to sizing errors and improving the customer experience on e-commerce platforms.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.3
• /
• pp.326-334
• /
• 2022

In this paper, we propose speech enhancement algorithm as a pre-processing for robust speech recognition in noisy environments. Auxiliary-function-based Independent Vector Analysis (AuxIVA) is performed with weighted covariance matrix using time-varying variances with scaling factor from target masks representing time-frequency contributions of target speech. The mask estimates can be obtained using Neural Network (NN) pre-trained for speech extraction or diffuseness using Coherence-to-Diffuse power Ratio (CDR) to find the direct sounds component of a target speech. In addition, outputs for omni-directional noise are closely chained by sharing the time-varying variances similarly to independent subspace analysis or IVA. The speech extraction method based on AuxIVA is also performed in Independent Low-Rank Matrix Analysis (ILRMA) framework by extending the Non-negative Matrix Factorization (NMF) for noise outputs to Non-negative Tensor Factorization (NTF) to maintain the inter-channel dependency in noise output channels. Experimental results on the CHiME-4 datasets demonstrate the effectiveness of the presented algorithms.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.2
• /
• pp.123-129
• /
• 2017

This paper proposes a new nonnegative matrix factorization (NMF) based direction-of-arrival (DOA) estimation method for multiple sound sources using a dual microphone array. First of all, sound signals coming from the dual microphone array are segmented into consecutive analysis frames, and a steered-response power phase transform (SRP-PHAT) beamformer is applied to each frame so that stereo signals of each frame are represented in a time-direction domain. The time-direction outputs of SRP-PHAT are stored for a pre-defined number of frames, which is referred to as a time-direction block. Next, In order to estimate DOAs robust to noise, each time-direction block is normalized along the time by using a block subtraction technique. After that, an unsupervised NMF method is applied to the normalized time-direction block in order to cluster the directions of each sound source in a multiple sound source environments. In particular, the activation and basis matrices are used to estimate the number of sound sources and their DOAs, respectively. The DOA estimation performance of the proposed method is evaluated by measuring a mean absolute error (MAE) and the standard deviation of errors between the oracle and estimated DOAs under a three source condition, where the sources are located in [$-35{\circ}$, 5m], [$12{\circ}$, 4m], and [$38{\circ}$, 4.m] from the dual microphone array. It is shown from the experiment that the proposed method could relatively reduce MAE by 56.83%, compared to a conventional SRP-PHAT based DOA estimation method.

• Journal of the Korea Society for Simulation
• /
• v.32 no.1
• /
• pp.35-44
• /
• 2023

In recent, there are many research cases that analyze trends or research trends with text analysis. When collecting documents by searching for keywords in abbreviations for data analysis, it is necessary to disambiguate abbreviations. In many studies, documents are classified by hand-work reading the data one by one to find the data necessary for the study. Most of the studies to disambiguate abbreviations are studies that clarify the meaning of words and use supervised learning. The previous method to disambiguate abbreviation is not suitable for classification studies of documents looking for research data from abbreviation search documents, and related studies are also insufficient. This paper proposes a method of semi-automatically classifying documents collected by abbreviations by going topic modeling with Non-Negative Matrix Factorization, an unsupervised learning method, in the data pre-processing step. To verify the proposed method, papers were collected from academic DB with the abbreviation 'MSA'. The proposed method found 316 papers related to Micro Services Architecture in 1,401 papers. The document classification accuracy of the proposed method was measured at 92.36%. It is expected that the proposed method can reduce the researcher's time and cost due to hand work.

• The Journal of the Acoustical Society of Korea
• /
• v.35 no.4
• /
• pp.303-309
• /
• 2016

This paper proposes a novel approach to the problem of audio signal clustering using a stacked autoencoder. The proposed stacked autoencoder learns an efficient representation for the input signal, enables clustering constituent signals with similar characteristics, and therefore the original sources can be separated based on the clustering results. STFT (Short-Time Fourier Transform) is performed to extract time-frequency spectrum, and rectangular windows at all the possible locations are used as input values to the autoencoder. The outputs at the middle, encoding layer, are used to cluster the rectangular windows and the original sources are separated by the Wiener filters derived from the clustering results. Source separation experiments were carried out in comparison to the conventional NMF (Non-negative Matrix Factorization), and the estimated sources by the proposed method well represent the characteristics of the orignal sources as shown in the time-frequency representation.