Moon-Hyun Kim

A New Diverse AdaBoost Classifier

AdaBoost is one of the most popular algorithms to construct a strong classifier with linear combination of member classifiers. The member classifiers are selected to minimize the errors in each iteration step during training process. AdaBoost provides very simple and useful method to generate ensemble classifiers. The performance of the ensemble depends on the diversity among the member classifiers as well as the performance of each member classifiers. However the existing AdaBoost algorithms are focused on error minimization problems. In this paper, we propose a noble method to inject diversity into the AdaBoost process to improve the performance of the AdaBoost classifiers. The proposed Diverse AdaBoost algorithm outperforms Gentle AdaBoost algorithm, because of the injected diversity. Our research contributes to the method designing optimized ensemble classifiers with diversity.

Automated measurement of crowd density based on edge detection and optical flow

In metropolitan area, the use of analog CCTV surveillance system has steadily increased to detect and prevent crimes, terror, big persons and cars crowded in road that CCTV surveillance systems have long been employed in metropolitan area. It is necessary to equip intelligent surveillance systems, which provide an efficient solution for the problems because the cities become bigger than now in population. Therefore, a method that can efficiently measure crowd density is needed. We present an efficient method that estimates information of crowd movement by using the optical flow and the numbers of moving objects by using the numbers of edge pixels. Finally, we will classify the crowd density using the two estimated information according to five levels of crowd density.

Estimation of Crowd Density in Public Areas Based on Neural Network

There are nowadays strong demands for intelligent surveillance systems, which can infer or understand more complex behavior. The application of crowd density estimation methods could lead to a better understanding of crowd behavior, improved design of the built environment, and increased pedestrian safety. In this paper, we propose a new crowd density estimation method, which aims at estimating not only a moving crowd, but also a stationary crowd, using images captured from surveillance cameras situated in various public locations. The crowd density of the moving people is measured, based on the moving area during a specified time period. The moving area is defined as the area where the magnitude of the accumulated optical flow exceeds a predefined threshold. In contrast, the stationary crowd density is estimated from the coarseness of textures, under the assumption that each person can be regarded as a textural unit. A multilayer neural network is designed, to classify crowd density levels into 5 classes. Finally, the proposed method is experimented with PETS 2009 and the platform of Gangnam subway station image sequences.

An Application of Detecting Plagiarism using Dynamic Incremental Comparison Method

At present, a lot of information that is provided online is actually being plagiarized or illegally copied. Specifically, it is very tricky to identify some plagiarism from tremendous amount of information because the original sentences can be simply restructured or replaced with similar words, which would make them look different from original sentences. This means that managing and protecting the knowledge start to be regarded as important, though it is important to create the knowledge through the investment and efforts. This dissertation tries to suggest new method and theory that would be instrumental in effectively detecting any infringement on and plagiarism of intellectual property of others. Dynamic incremental comparison method, a method which was developed by this research to detect plagiarism of document, focuses on realizing a system that can detect plagiarized documents and parts efficiently, accurately and immediately by creating positive and various detectors

Comparison of clustering methods for MLP-based speaker verification

This paper compares two clustering methods: SOM, and a graph-based clustering technique , for text-independent speaker verification. The focus of comparison is given to the distribution characteristics of representative frames for each cluster, to the use of processing time of clustering and MLP learning, and to verification performance. Simulation results show that the graph-based technique produces better verification performance than SOM. Other statistics are collected to explain significant difference in MLP learning time with each clustering method. This experiment suggests that there is a best match between a classifier and a clustering method for a given application.

Multi-Sensor Signal based Situation Recognition with Bayesian Networks

In this paper, we propose an intelligent situation recognition model by collecting and analyzing multiple sensor signals. Multiple sensor signals are collected for fixed time window. A training set of collected sensor data for each situation is provided to K2-learning algorithm to generate Bayesian networks representing causal relationship between sensors for the situation. Statistical characteristics of sensor values and topological characteristics of generated graphs are learned for each situation. A neural network is designed to classify the current situation based on the extracted features from collected multiple sensor values. The proposed method is implemented and tested with UCI machine learning repository data.

A Statistical Approach to Robust Background Subtraction for Urban Traffic Video

In this paper we present a robust method for background subtraction from a fixed camera in video surveillance system. The background subtraction is an important part of object tracking and many algorithms have been proposed for decades. Mixture of Gaussian for those in this paper is very famous and used widely. We present the robust method that can adapt the background model to various situations. We have to detect not only moving objects but also stopped objects, but this detecting problem have not been solved in the previous research. To solve this problem, we present an efficient adaptive Mixture of Gaussian Model in urban transit. The parameter should be adapted in various situations. We train the model and get the adaptive parameter by using the time gap between moving and stopped objects. This model can be applied to the real-time application. We demonstrate and evaluate our proposed method with urban traffic sequences.

Text independent speaker verification using modular neural network

This work addresses the data balancing problem of the existing neural network based speaker verification methods, and proposes new method using modular neural network. In this method, each expert network is trained with the balanced number of genuine speaker data and imposter speaker data. In our experiments, we obtained high performance results for the unknown imposter speakers. High performance and the modular nature of the proposed method enables building a large scalable speaker verification system.

Measurement of the Crowd Density in Outdoor Using Neural Network

The population growth along with the urbanization, has caused more problems in many public areas, such as subway airport terminals, hospital, etc. Many surveillance systems have been installed in the public areas, but not all of those can be monitored in real-time, because the operators that observe the monitors are very small compared with the number of the monitors. For example, the observer can miss some crucial accidents or detect after considerable delays. Thus, intelligent surveillance system for preventing the accidents are needed, such as Intelligent Surveillance Systems. in this paper, we propose a new crowd density estimation method which aims at estimating moving crowd using images from surveillance cameras situated in outdoor locations. The moving crowd is estimated from the area where using optical flow. The edge information is also used as feature to measure the crowd density, so we improve the accuracy of estimation of crowd density. A multilayer neural network is designed to classify crowd density into 5 classes. Finally the proposed method is experimented with PETS 2009 images.

Korean documents copy detection based on ferret

With the development of electronic documents, plagiarism is rapidly increasing and, given the difficulty of manual detection, need for plagiarism detection systems to help protect intellectual property has emerged. Many content-based detection systems have been developed and are actually used in some foreign countries, but they are still insufficient for documents in Korean. In particular, the high variance of Hangul makes the development of detection systems more difficult. This study proposes a Hangul document detection method based on Ferrets trigrams. Ferret only considered the frequency of trigram matches as a way to detect similarity, but in this study the system is developed further by weighting results depending on the degree of trigram match, thereby improving the accuracy of similarity detection.