Search

Kinetic typography is a technology based on the changes of the color of the character, size, and location that delivers appreciation of the beautiful and creativity. This research’s main purpose is in to use the depth-camera to make natural interaction between users and kinetic typography. We used motion recognition sensor in order to track user’s skeleton informations to synchronize with typography. Kinetic typography does not only deliver simple information, it also allows high dimensional communication such as emotional expression.

In human-agent interactions, attention sharing plays a key role in understanding other’s intention without explicit verbal explanation. Deep learning algorithms are recently used to model these interactions in a complex real world environment. In this paper we propose a deep learning based intention estimation and recommendation system by understanding humans attention based on their gestures. Action-object affordances are modeled using stacked auto-encoder, which represents the relationships between actions and objects. Intention estimation and object recommendation system according to human intention is implemented based on an affordance model. Experimental result demonstrates meaningful intention estimation and recommendation performance in the real-world scenarios.

Deep learning methods aims at learning features automatically at multiple levels that allow the system to learn complex functions mapping the input to the output directly from data. This ability to automatically learn powerful features will become increasingly important as the amount of data and range of applications to machine learning methods continues to grow. In this context we propose a deep architecture model using Support Vector Machine (SVM) which has inherent ability to select data points important for classification with good generalization capabilities. Since SVM can effectively discriminate features, we used support vectors with kernel as non-linear discriminant features for classification. By stacking SVMs in to multiple layers, we can obtain deep features without extra feature engineering steps and get robust recognition accuracy. Experimental results show that the proposed method improves generalization performance on Wisconsin Breast Cancer dataset.

Existing drawing robots draw pictures using robot-arm based on coordinates extracted from image processing. However, they draws pictures by presenting contours and brightness, they lacked the technique of stepped representation of brightness and due to that, resulting pictures differed a lot compared to its original version. This paper proposes art robot system that is capable of detection and extraction of brightness in images, along with dynamic representation of brightness lines that change with accordance to the value of brightness level. Proposed robot system analyzes and classifies the brightness by its density and based on this brightness levels, brightness and shade is drawn by repeated reproducing of overlapped lines. This representation of brightness method can draw pictures with various types of brightness by adjusting the number of brightness lines and leveling of brightness.

In self-organizing network, detecting cluster or group structure reduces a complexity of the network consisted of large datasets, which is important to manage and understand components of the network. Especially, in this network, detecting accurately clusters is an important technique to realize a service of group action. Therefore, we propose a discovery technique of overlapping cluster in self-organizing network. Proposed method performs hierarchical clustering to detect and identify groups. To achieve this, we make a hierarchy structure by comparing similarity between two edges sharing one node. All of edges in network are represented with tree; these are grouped by specific modularity which is a boundary to cut groups.

Nowadays social network systems become essential ingredients in daily life. Meanwhile, with the booming of smart mobile terminals, mobile social network catches the attention of world. However, most widely used mobile social network systems are focused on perceiving neighbors with common interests, which is just a cross-section of social network. This paper proposes Context-driven Mobile Social Network Discovery System, a system that enables users to have an overall view of nearby social network resources and provides a platform for discovering and recommending nearby social network, taking an active part in social activities and establishing more useful relationships in social communities.

As home network is expanding into ubiquitous computing environment and lots of private information is accessible, it is required to protect home network system from illegal accesses and security threats in open network. In general deployment of home network, a secure home gateway is located at the boarder of each home and regarded to be a core entity providing services and controlling traffic. So in this paper, we propose a security system guaranteeing security, availability, and reliability based on the secure home gateway and describe our implementation on home network including authentication, authorization, and security policy.

As a number of home network services are available and home network is expanding into ubiquitous computing environment, we need to protect home network system from illegal accesses and a variety of threats. Home network is exposed to various cyber attacks of Internet, involves hacking, malicious codes, worms, viruses, DoS attacks, and eavesdropping since it is connected to Internet. So in this paper, we propose a home network security framework for guaranteeing reliability and availability including authentication, authorization and security policy system.

In ubiquitous computing environment, people carrying their mobile devices (eg., mobile phone, PDA, embedded devices) expect to access locally hosted services or resources anytime, anywhere. These mobile devices have restricted capabilities and security supports. Traditional security management systems used definite access control policies for each role or user in each domain server or agent. But in ubiquitous environment, it is hard to specify authorization policies for mobile users and it is inflexible and unavailable for security management of users or mobile devices. To solve these problems, we need trust-based management mechanism as a reference to security management systems. Trust model contains trust relationship and calculation of trust value. Experiences and recommendations are the factors to calculate trust value. In this paper, we design a trust model to calculate trust value and a trust management architecture which can be running in various domain servers and mobile devices.

There are a variety of AV devices and multimedia content in the ubiquitous home and the sharing of content needs to have high quality of transfer by using optimized transport protocol for the each content. This paper presents the design and implementation of our proposed scheme that is based on UPnP AV framework which can support various transport protocol transparently. We validated and evaluated our proposed scheme and experimental results show that it can stream content effectively.

Wireless Local Area Network (WLAN) is susceptible to security provisioning in spite of the solutions such as the Wired Equivalent Protocol (WEP) or IEEE 802.1x This paper proposes an integrated scheme for intrusion detection in WLAN systems The proposed scheme operates with one or more Gathering Agents (GAs) and a Master Server (MS) Each GA is used to get security information by collecting the frame packets in WLAN, whereas the MS is purposed to detect and prevent the various attacks by analyzing the packets in the WLAN systems A detection engine contained in the MS employs OUI list matching for detection of MAC spoofing attacks, sequence number analysis for man-in-the-middle attacks, and Finite State Machine (FSM) analysis for Denial-of-Service (DoS) attacks By experiments, it is shown that the proposed scheme could effectively detect and prevent the various attacks that could possibly be done in the WLAN systems.

This paper proposes a dynamic IP address configuration (DAC) scheme for mSCTP handover, which exploits the information from the link layer to support SCTP handover between heterogeneous access networks. The proposed DAC scheme enables a mobile terminal to automatically add a new IP address in the newly visited region, change the primary path, and delete the old IP address so as to support the mSCTP handover. For experimental analysis of mSCTP handover, we consider the handover scenarios for 3G-BWA and 3G-WLAN over Linux platforms. From experimental results, it is shown that the throughput and handover latency of the mSCTP handover would be affected by the timing of primary-change and the overlapping period

Stream Control Transmission Protocol (SCTP) is a new end-to-end transport protocol that is featured ‘multi-homing.' The mSCTP (mobile SCTP) is defined as SCTP with the capability of dynamic address reconfiguration. This paper describes a framework of mSCTP handover for supporting vertical handover between heterogeneous IP networks such as WLAN and 3G Cellular systems. We show some experimental results of the mSCTP vertical handover on Linux platforms. From the experimental results, we see that the handover latency of mSCTP depends on Round Trip Time (RTT) between two SCTP endpoints, possibly with the handover latency of 1 second below.

We describe an implementation of an efficient player for MPEG-4 contents on a mobile device. The player uses 3 adaptation methods in order to support high efficiency when it plays MPEG-4 contents. The adaptation methods are composed of a physical adaptation, an event adaptation, and a resource adaptation. The first adaptation resolves the physical difference between the authoring environment and playback environment. The second adaptation resolves some events that cannot support on mobile devices. The last adaptation resolves the load for restricted resource on mobile device. In view of the results of the performance analysis on the player, we could find that the player showed an efficient playback of MPEG-4 contents on a mobile device. The applications for the player are as follows; mobile games, a car navigator, distance learning, etc.

MPEG-4 supports object-based interactive multimedia applications. To compose complex multimedia scene, MPEG-4 supports BIFS and XMT as scene description. For interoperability of the MPEG-4 content and object-based rendering information management, this paper proposes the content analyzer that supports multiple scene description. In order to improve content management’s efficiency, the content analyzer provides module optimization from scene description’s parsing to scene information management. This analyzer supports exact analysis of the scene description and object-based core information management. And it can be applied to XML-based heterogeneous players.

Stream Control Transmission Protocol (SCTP) is a new end-to-end transport protocol, which can be used to support the mobility of mobile terminals. This paper describes a framework of SCTP handover and analyzes the handover latency for the single-homing mobile terminals. We then show the experimental results of the SCTP handover in terms of the handover latency and throughput for the two different movement patterns: linear and crossover patterns. For the linear movement pattern, it is shown that the SCTP handover latency may severely depend on the handover delay at the underlying link layer. In the case of the crossover movement pattern, we see that the throughput of the data transmission could be degraded, as the crossover movements occur more frequently.

MPEG-4 supports dynamic scene composition through various interactions such as adding/removing/replacing objects and changing object’s properties. MPEG-4 content can express rich meanings through various dynamic scene composition by object-based, and can be played to mobile or desktop environment. Therefore, the MPEG-4 player must effectively play according authoring purpose and meaning, and compose dynamic scene in any device. This paper proposes an adaptive scene compositor model. It is optimized for dynamic scene composition of MPEG-4 content in mobile device. This model supports exact analysis of the scene description, core information extraction for rendering through extraction rule table and object-based information management for object’s reusability and flexibility.

We propose a transforming method of MPEG-4 contents in order to present the contents on PDA devices in this paper. The method reconstructs the mp4 file according to transforming a scene tree in MPEG-4 contents when the contents are authored. The method reduces the size of each object in a scene of the content for presenting it efficiently on the small interface of a PDA device, and transforms visual objects in the scene into geometry objects in order to reduce initial loading time and a size of the contents. And an original object is presented on a PDA device when the user clicks the substituted geometry object. The method was applied to a conventional authoring tool, so we could find that the method showed an efficient presentation of the contents on a PDA device.

A common breach of network security is the class of attacks called Worm-virus. This paper proposes a language called Triton whose goal is to efficiently and effectively safeguard large network security for management system. We represent abstract language that supports various functions. This high-level language abstracts the control behavior of the network nodes, which has various setting-up methodology. We shall describe the feature of Triton language and give some example of the preliminary Triton implementation on the test-bed.

Since current internet intruders conceal their real identity by distributed or disguised attacks, it is not easy to deal with intruders properly only with an ex post facto chase. Therefore, it needs to trace the intruder in real time. Existing real-time intruder tracing systems has a spatial restriction. The security domain remains unchanged if there is no system security officer's intervention after installing the tracing system. It is impossible to respond to an attack which is done out of the security domain. This paper proposes self-replication mechanism, a new approach to real-time intruder tracing, minimizing a spatial limitation of traceable domain. The real-time tracing supports prompt response to the intrusion, detection of target host and laundering hosts. It also enhances the possibility of intruder identification. Collected data during the real-time tracing can be used to generate a hacking scenario database and can be used as legal evidence.

Due to the development of wireless internet and an increasing number of internet users, transferring and receiving errorless data in real-time can be the most important method to guarantee the QoS (Quality of Service) of internet. Convolutional encoding and Viterbi decoding are the widely used techniques to enhance the performance of BER (bit error rate) in the application area such as satellite communications systems. As a method to enhance the QoS of internet, a new DSP architecture that can effectively materialize the Viterbi algorithm, one of the algorithms that can correct errors during data transfer, is introduced in this paper. A new architecture and a new instruction set, which can handle the Viterbi algorithm faster, and simplify the Euclidean distance calculation, are defined. The performance assessment result shows that the proposed DSP can execute the Viterbi algorithm faster than other DSPs. Using 0.18 µm CMOS technology, the new DSP operates in 100 MHz, and consumes 218 µA/MHz.

MPEG-4 is an ISO/IEC standard which defines a multimedia system for communicating interactive scenes. An MPEG-4 audio-visual scene is composed of various types of media data that are represented as individual objects. It includes a scene description that refers to the spatio-temporal specifications and behaviors of the individual objects. Particularly in the presentation of an MPEG-4 scene, object-based user interaction is possible within the limit set defined by the author. In this paper, we propose an MPEG-4 scene authoring tool which provides facilitative tools for composing interactive behaviors of the objects. The authoring tool presents a graphical interface for building a temporal scenario of MPEG-4 scenes. It also proposes new means of graphically representing an interactive scenario.