• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.12C
• /
• pp.1200-1208
• /
• 2003

Affective computing plays an important role in intelligent Human Computer Interactions(HCI). To detect emotional events, it is desirable to construct a computing model for extracting emotion related features from video. In this paper, we propose a statistical model based on the probabilistic distribution of low level features in video shots. The proposed method extracts low level features from video shots and then from a GMM(Gaussian Mixture Model) for them to detect emotional shots. As low level features, we use color, camera motion and sequence of shot lengths. The features can be modeled as a GMM by using EM(Expectation Maximization) algorithm and the relations between time and emotions are estimated by MLE(Maximum Likelihood Estimation). Finally, the two statistical models are combined together using Bayesian framework to detect emotional events in video.

• Journal of Digital Convergence
• /
• v.13 no.2
• /
• pp.135-140
• /
• 2015

General CHMM vocabulary recognition system is model observation probability for vocabulary recognition of recognition rate's low. Used as the limiting unit is applied only to some problem in the phoneme model. Also, they have a problem that does not conform to the needs of the search range to meaning of the words in the vocabulary. Performs a phoneme recognition using GMM to improve these problems. We solve the problem according to the limited search words characterized by an improved k-means algorithm. Measure the effectiveness represented by the accuracy and reproducibility as compared to conventional system performance experiments. Performance test results accuracy is 83%p, and recall is 67%p.

• International Journal of Advanced Culture Technology
• /
• v.10 no.2
• /
• pp.252-259
• /
• 2022

Artificial intelligence is used in fusion with image processing techniques using cameras. Image processing technology is a technology that processes objects in an image received from a camera in real time, and is used in various fields such as security monitoring and medical image analysis. If such image processing reduces the accuracy of recognition, providing incorrect information to medical image analysis, security monitoring, etc. may cause serious problems. Therefore, this paper uses a mixture of YOLOv4-tiny model and image processing algorithm and uses the COCO dataset for learning. The image processing algorithm performs five image processing methods such as normalization, Gaussian distribution, Otsu algorithm, equalization, and gradient operation. For RGB images, three image processing methods are performed: equalization, Gaussian blur, and gamma correction proceed. Among the nine algorithms applied in this paper, the Equalization and Gaussian Blur model showed the highest object detection accuracy of 96%, and the gamma correction (RGB environment) model showed the highest object detection rate of 89% outdoors (daytime). The image binarization model showed the highest object detection rate at 89% outdoors (night).

• Journal of the Institute of Convergence Signal Processing
• /
• v.3 no.1
• /
• pp.8-14
• /
• 2002

In this paper, we realized a real-time text-independent speaker recognition system using gaussian mixture model, and applied frame level likelihood normalization method which shows its effects in verification system. The system has three parts as front-end, training, recognition. In front-end part, cepstral mean normalization and silence removal method were applied to consider speaker's speaking variations. In training, gaussian mixture model was used for speaker's acoustic feature modeling, and maximum likelihood estimation was used for GMM parameter optimization. In recognition, likelihood score was calculated with speaker models and test data at frame level. As test sentences, we used text-independent sentences. ETRI 445 and KLE 452 database were used for training and test, and cepstrum coefficient and regressive coefficient were used as feature parameters. The experiment results show that the frame-level likelihood method's recognition result is higher than conventional method's, independently the number of registered speakers.

• Economic and Environmental Geology
• /
• v.55 no.4
• /
• pp.353-366
• /
• 2022

Spatial estimation of geoscience data (geo-data) is challenging due to spatial heterogeneity, data scarcity, and high dimensionality. A novel spatial estimation method is needed to consider the characteristics of geo-data. In this study, we proposed the application of Gaussian Mixture Model (GMM) among machine learning algorithms with multivariate data for robust spatial predictions. The performance of the proposed approach was tested through soil chemical concentration data from a former smelting area. The concentrations of As and Pb determined by ex-situ ICP-AES were the primary variables to be interpolated, while the other metal concentrations by ICP-AES and all data determined by in-situ portable X-ray fluorescence (PXRF) were used as auxiliary variables in GMM and ordinary cokriging (OCK). Among the multidimensional auxiliary variables, important variables were selected using a variable selection method based on the random forest. The results of GMM with important multivariate auxiliary data decreased the root mean-squared error (RMSE) down to 0.11 for As and 0.33 for Pb and increased the correlations (r) up to 0.31 for As and 0.46 for Pb compared to those from ordinary kriging and OCK using univariate or bivariate data. The use of GMM improved the performance of spatial interpretation of anthropogenic metals in soil. The multivariate spatial approach can be applied to understand complex and heterogeneous geological and geochemical features.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.1
• /
• pp.64-68
• /
• 2020

In a conventional speech recognition system using Gaussian Mixture Model-Hidden Markov Model (GMM-HMM), the cepstral feature normalization method based on pole filtering was effective in improving the performance of recognition of short utterances in noisy environments. In this paper, the usefulness of this method for the state-of-the-art speech recognition system using Deep Neural Network (DNN) is examined. Experimental results on AURORA 2 DB show that the cepstral mean and variance normalization based on pole filtering improves the recognition performance of very short utterances compared to that without pole filtering, especially when there is a large mismatch between the training and test conditions.

• Smart Structures and Systems
• /
• v.5 no.1
• /
• pp.95-117
• /
• 2009

In this paper, the time series based damage detection algorithms developed by Nair, et al. (2006) and Nair and Kiremidjian (2007) are applied to the benchmark experimental data from the National Center for Research on Earthquake Engineering (NCREE) in Taipei, Taiwan. Both acceleration and strain data are analyzed. The data are modeled as autoregressive (AR) processes, and damage sensitive features (DSF) and feature vectors are defined in terms of the first three AR coefficients. In the first algorithm developed by Nair, et al. (2006), hypothesis tests using the t-statistic are applied to evaluate the damaged state. A damage measure (DM) is defined to measure the damage extent. The results show that the DSF's from the acceleration data can detect damage while the DSF from the strain data can be used to localize the damage. The DM can be used for damage quantification. In the second algorithm developed by Nair and Kiremidjian (2007) a Gaussian Mixture Model (GMM) is used to model the feature vector, and the Mahalanobis distance is defined to measure damage extent. Additional distance measures are defined and applied in this paper to quantify damage. The results show that damage measures can be used to detect, quantify, and localize the damage for the high intensity and the bidirectional loading cases.

• Journal of information and communication convergence engineering
• /
• v.11 no.2
• /
• pp.95-102
• /
• 2013

This paper presents a stochastic driver behavior modeling framework which takes into account both individual and general driving characteristics as one aggregate model. Patterns of individual driving styles are modeled using a Dirichlet process mixture model, as a non-parametric Bayesian approach which automatically selects the optimal number of model components to fit sparse observations of each particular driver's behavior. In addition, general or background driving patterns are also captured with a Gaussian mixture model using a reasonably large amount of development data from several drivers. By combining both probability distributions, the aggregate driver-dependent model can better emphasize driving characteristics of each particular driver, while also backing off to exploit general driving behavior in cases of unseen/unmatched parameter spaces from individual training observations. The proposed driver behavior model was employed to anticipate pedal operation behavior during car-following maneuvers involving several drivers on the road. The experimental results showed advantages of the combined model over the model adaptation approach.

• ETRI Journal
• /
• v.38 no.3
• /
• pp.579-588
• /
• 2016

To improve the ability to determine a vehicle's movement information even in a challenging environment, a hybrid approach called non-Gaussian square rootunscented particle filtering (nGSR-UPF) is presented. This approach combines a square root-unscented Kalman filter (SR-UKF) and a particle filter (PF) to determinate the vehicle state where measurement noises are taken as a finite Gaussian kernel mixture and are approximated using a sparse Gaussian kernel density estimation method. During an outage-free GPS period, the updated mean and covariance, computed using SR-UKF, are estimated based on a GPS observation update. During a complete GPS outage, nGSR-UPF operates in prediction mode. Indeed, because the inertial sensors used suffer from a large drift in this case, SR-UKF-based importance density is then responsible for shifting the weighted particles toward the high-likelihood regions to improve the accuracy of the vehicle state. The proposed method is compared with some existing estimation methods and the experiment results prove that nGSR-UPF is the most accurate during both outage-free and complete-outage GPS periods.

• Journal of Korea Multimedia Society
• /
• v.15 no.6
• /
• pp.748-760
• /
• 2012

In case brightness distribution is concentrated in a region, it is difficult to classify the image features. To solve this problem, we apply global histogram equalization and local histogram equalization to images. In case of global histogram equalization, it can be too bright or dark because it doesn't consider the density of brightness distribution. Thus, it is difficult to enhance the local contrast in the images. In case of local histogram equalization, it can produce unexpected blocks in the images. In order to enhance the contrast in the images, this paper proposes a local histogram equalization based on the Gaussian Mixture Models(GMMs) in regions of histogram. Mean and variance parameters in each regions is updated EM-algorithm repeatedly and then ranges of equalization on each regions. The experimental results performed with image of various contrasts show that the proposed algorithm is better than the global histogram equalization.