• Journal of Intelligence and Information Systems
• /
• v.25 no.4
• /
• pp.53-65
• /
• 2019

Object tracking is one of important steps to achieve video-based surveillance systems. Object tracking is considered as an essential task similar to object detection and recognition. In order to perform object tracking, various machine learning methods (e.g., least-squares, perceptron and support vector machine) can be applied for different designs of tracking systems. In general, generative methods (e.g., principal component analysis) were utilized due to its simplicity and effectiveness. However, the generative methods were only focused on modeling the target object. Due to this limitation, discriminative methods (e.g., binary classification) were adopted to distinguish the target object and the background. Among the machine learning methods for binary classification, total error rate minimization can be used as one of successful machine learning methods for binary classification. The total error rate minimization can achieve a global minimum due to a quadratic approximation to a step function while other methods (e.g., support vector machine) seek local minima using nonlinear functions (e.g., hinge loss function). Due to this quadratic approximation, the total error rate minimization could obtain appropriate properties in solving optimization problems for binary classification. However, this total error rate minimization was based on a batch mode setting. The batch mode setting can be limited to several applications under offline learning. Due to limited computing resources, offline learning could not handle large scale data sets. Compared to offline learning, online learning can update its solution without storing all training samples in learning process. Due to increment of large scale data sets, online learning becomes one of essential properties for various applications. Since object tracking needs to handle data samples in real time, online learning based total error rate minimization methods are necessary to efficiently address object tracking problems. Due to the need of the online learning, an online learning based total error rate minimization method was developed. However, an approximately reweighted technique was developed. Although the approximation technique is utilized, this online version of the total error rate minimization could achieve good performances in biometric applications. However, this method is assumed that the total error rate minimization can be asymptotically achieved when only the number of training samples is infinite. Although there is the assumption to achieve the total error rate minimization, the approximation issue can continuously accumulate learning errors according to increment of training samples. Due to this reason, the approximated online learning solution can then lead a wrong solution. The wrong solution can make significant errors when it is applied to surveillance systems. In this paper, we propose an exactly reweighted technique to recursively update the solution of the total error rate minimization in online learning manner. Compared to the approximately reweighted online total error rate minimization, an exactly reweighted online total error rate minimization is achieved. The proposed exact online learning method based on the total error rate minimization is then applied to object tracking problems. In our object tracking system, particle filtering is adopted. In particle filtering, our observation model is consisted of both generative and discriminative methods to leverage the advantages between generative and discriminative properties. In our experiments, our proposed object tracking system achieves promising performances on 8 public video sequences over competing object tracking systems. The paired t-test is also reported to evaluate its quality of the results. Our proposed online learning method can be extended under the deep learning architecture which can cover the shallow and deep networks. Moreover, online learning methods, that need the exact reweighting process, can use our proposed reweighting technique. In addition to object tracking, the proposed online learning method can be easily applied to object detection and recognition. Therefore, our proposed methods can contribute to online learning community and object tracking, detection and recognition communities.

• Journal of rehabilitation welfare engineering & assistive technology
• /
• v.11 no.4
• /
• pp.371-376
• /
• 2017

In this paper, we propose an algorithm that can recognize the pattern regardless of the sensor position when performing EMG pattern recognition using circular EMG system equipment. Fourteen features were extracted by using the data obtained by measuring the eight channel EMG signals of six motions for 1 second. In addition, 112 features extracted from 8 channels were analyzed to perform principal component analysis, and only the data with high influence was cut out to 8 input signals. All experiments were performed using k-NN classifier and data was verified using 5-fold cross validation. When learning data in machine learning, the results vary greatly depending on what data is learned. EMG Accuracy of 99.3% was confirmed when using the learning data used in the previous studies. However, even if the position of the sensor was changed by only 22.5 degrees, it was clearly dropped to 67.28% accuracy. The accuracy of the proposed method is 98% and the accuracy of the proposed method is about 98% even if the sensor position is changed. Using these results, it is expected that the convenience of the users using the circular EMG system can be greatly increased.

• The KIPS Transactions:PartB
• /
• v.16B no.1
• /
• pp.7-16
• /
• 2009

Some clustering algorithms have a problem that an image is over-segmented since both the spatial information between the segmented regions is not considered and the number of the clusters is defined in advance. Therefore, they are difficult to be applied to the applicable fields. This paper proposes the new post-processing methods, a reclassification of the inhomogeneous clusters and a region merging using Baysian algorithm, that improve the segmentation results of the clustering algorithms. The inhomogeneous cluster is firstly selected based on variance and between-class distance and it is then reclassified into the other clusters in the reclassification step. This reclassification is repeated until the optimal number determined by the minimum average within-class distance. And the similar regions are merged using Baysian algorithm based on Kullbeck-Leibler distance between the adjacent regions. So we can effectively solve the over-segmentation problem and the result can be applied to the applicable fields. Finally, we design a classification system for the segmented regions to validate the proposed method. The segmented regions are classified by SVM(Support Vector Machine) using the principal colors and the texture information of the segmented regions. In experiment, the proposed method showed the validity for various real-images and was effectively applied to the designed classification system.

• Journal of the Korea Society of Computer and Information
• /
• v.18 no.4
• /
• pp.45-55
• /
• 2013

In this study, we propose a tongue diagnosis system which detects the tongue from face image and divides the tongue area into six areas, and finally generates tongue fur ratio of each area. To detect the tongue area from face image, we use ASM as one of the active shape models. Detected tongue area is divided into six areas and the distribution of tongue coating of six areas is examined by SVM. For SVM, we use a 3-dimensional vector calculated by PCA from a 12-dimensional vector consisting of RGB, HSV, Lab, and Luv. As a result, we stably detected the tongue area using ASM. Furthermore, we recognized that PCA and SVM helped to raise the ratio of tongue coating detection.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.3
• /
• pp.54-60
• /
• 2010

The local appearance-based method is one of the face recognition methods that divides face image into small areas and extracts features from each area of face image using statistical analysis. It collects classification results of each area and decides identity of a face image using a voting scheme by integrating classification results of each area of a face image. The conventional local appearance-based method divides face images into small pieces and uses all the pieces in recognition process. In this paper, we propose a local appearance-based method that makes use of only the relatively important facial components. The proposed method detects the facial components such as eyes, nose and mouth that differs much from person to person. In doing so, the proposed method detects exact locations of facial components using support vector machines (SVM). Based on the detected facial components, a number of small images that contain the facial parts are constructed. Then it extracts features from each facial component image using principal components analysis (PCA). We compared the performance of the proposed method to those of the conventional methods. The results show that the proposed method outperforms the conventional local appearance-based method while preserving the advantages of the conventional local appearance-based method.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.20 no.1
• /
• pp.33-39
• /
• 2017

In this paper, an effective feature which is capable of classifying targets among the detections obtained from 2D range-bearing maps generated in active sonar environments is proposed. Most conventional approaches for target classification with the 2D maps have considered magnitude of peak and statistical features of the area surrounding the peak. To improve the classification performance, HOG(Histogram of Gradient) feature, which is popular for their robustness in the image textures analysis is applied. In order to classify the target signal, SVM(Support Vector Machine) method with reduced HOG feature by the PCA(Principal Component Analysis) algorithm is incorporated. The various simulations are conducted with the real clutter signal data and the synthesized target signal data. According to the simulated results, the proposed method considering HOG feature is claimed to be effective when classifying the active sonar target compared to the conventional methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.4 no.1
• /
• pp.25-44
• /
• 2010

With the increasing popularity of mobile devices, it has become necessary to protect private information and content in these devices. Face recognition has been favored over conventional passwords or security keys, because it can be easily implemented using a built-in camera, while providing user convenience. However, because mobile devices can be used both indoors and outdoors, there can be many illumination changes, which can reduce the accuracy of face recognition. Therefore, we propose a new face recognition method on a mobile device robust to illumination variations. This research makes the following four original contributions. First, we compared the performance of face recognition with illumination variations on mobile devices for several illumination normalization procedures suitable for mobile devices with low processing power. These include the Retinex filter, histogram equalization and histogram stretching. Second, we compared the performance for global and local methods of face recognition such as PCA (Principal Component Analysis), LNMF (Local Non-negative Matrix Factorization) and LBP (Local Binary Pattern) using an integer-based kernel suitable for mobile devices having low processing power. Third, the characteristics of each method according to the illumination va iations are analyzed. Fourth, we use two matching scores for several methods of illumination normalization, Retinex and histogram stretching, which show the best and $2^{nd}$ best performances, respectively. These are used as the inputs of an SVM (Support Vector Machine) classifier, which can increase the accuracy of face recognition. Experimental results with two databases (data collected by a mobile device and the AR database) showed that the accuracy of face recognition achieved by the proposed method was superior to that of other methods.

• Food Science of Animal Resources
• /
• v.38 no.2
• /
• pp.362-375
• /
• 2018

This study proposed a rapid microscopic examination method for pork freshness evaluation by using the self-assembled hyperspectral microscopic imaging (HMI) system with the help of feature extraction algorithm and pattern recognition methods. Pork samples were stored for different days ranging from 0 to 5 days and the freshness of samples was divided into three levels which were determined by total volatile basic nitrogen (TVB-N) content. Meanwhile, hyperspectral microscopic images of samples were acquired by HMI system and processed by the following steps for the further analysis. Firstly, characteristic hyperspectral microscopic images were extracted by using principal component analysis (PCA) and then texture features were selected based on the gray level co-occurrence matrix (GLCM). Next, features data were reduced dimensionality by fisher discriminant analysis (FDA) for further building classification model. Finally, compared with linear discriminant analysis (LDA) model and support vector machine (SVM) model, good back propagation artificial neural network (BP-ANN) model obtained the best freshness classification with a 100 % accuracy rating based on the extracted data. The results confirm that the fabricated HMI system combined with multivariate algorithms has ability to evaluate the fresh degree of pork accurately in the microscopic level, which plays an important role in animal food quality control.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.12 no.1
• /
• pp.131-140
• /
• 1992

For a real-time flood forecasting and reservoir control problem of a multipurpose dam, the online acquisition of hydro-meteorological data and computerized analysis of the acquired data are absolutely necessary for the prompt decision of reservoir discharges which can minimize the possible damages and simultaneously maximize the utilization of the runoff. By introducing a man-machine interface such as condensed color graphics of the analyzed results, it is much easier and faster to transform the information to the decision maker who can decide the reservoir discharge. The newly developed PC-REFCON, which represents the PC based real-time flood forecasting and reservoir control, can easily handle the above problems by adopting a innovative decision support system. The system has three principal components of, a data base subsystem which acquires and manages real-time data, a model subsystem which forecasts the flood runoff and simulates the reservoir operation, and a dialogue subsystem which helps decision maker and system engineers using various graphics and tables with renovative methodologies. The developed PC-REFCON will be utilized from the coming Summer of 1992 for the flood control of all the nine multipurpose reservoirs in Korea.

• The Korean Journal of Applied Statistics
• /
• v.31 no.5
• /
• pp.539-554
• /
• 2018

Effective interaction between user and device is considered an important ability of IoT devices. For some applications, it is necessary to recognize human facial expressions in real time and make accurate judgments in order to respond to situations correctly. Therefore, many researches on facial image analysis have been preceded in order to construct a more accurate and faster recognition system. In this study, we constructed an automatic recognition system for facial expressions through two steps - a facial recognition step and a classification step. We compared various models with different sets of data with pixel information, landmark coordinates, Euclidean distances among landmark points, and arctangent angles. We found a fast and efficient prediction model with only 30 principal components of face landmark information. We applied several prediction models, that included linear discriminant analysis (LDA), random forests, support vector machine (SVM), and bagging; consequently, an SVM model gives the best result. The LDA model gives the second best prediction accuracy but it can fit and predict data faster than SVM and other methods. Finally, we compared our method to Microsoft Azure Emotion API and Convolution Neural Network (CNN). Our method gives a very competitive result.