電気学会論文誌C(電子・情報・システム部門誌)
Online ISSN : 1348-8155
Print ISSN : 0385-4221
ISSN-L : 0385-4221
121 巻, 10 号
選択された号の論文の24件中1~24を表示しています
  • 齊藤 英雄
    2001 年 121 巻 10 号 p. 1493-1499
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
  • 斉藤 文彦
    2001 年 121 巻 10 号 p. 1500-1507
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    The template matching is known as a general-purposed image processing method. However, the target image area to be searched must have the right position and the same scale as the template image. This paper proposes a method to search the target image area efficiently that has a free location, inclination and scale in an objective image. The genetic algorithm is applied to both the template registration process and the matching process to perform the efficient search. In the template registration process, only important pixels to perform the reliable matching are extracted. In the matching process, the parameters that represent the pose of the target image area are searched. The experimental results show that the proper target image areas with free locations, inclinations and scales were searched successfully and the averaged processing time was reached to 1.1 second.
  • Fumihiko Saitoh
    2001 年 121 巻 10 号 p. 1508-1515
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    This paper proposes a method to obtain a good contrast image by optimizing the number of local areas and by changing the positions of the boundaries for local contrast enhancement according to the conditions of an objective gray-scaled image. The genetic algorithm is applied to determine the optimal image separation. A chromosome is represented by a pair of arrays that show the positions of the boundaries to separate an image. The fitness of an individual is evaluated by the sharpness of the image that is measured by the delta-histograms of the enhanced image and the blurred image. The experimental results show that the natural enhanced images with few noises were generated by the proposed method in comparison with the conventional methods.
  • 中島 慶人, 伊藤 憲彦
    2001 年 121 巻 10 号 p. 1516-1523
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    This paper proposes a method of object recognition and their position-pose estimation by image processing under the constant illumination condition. The proposed method is a combination of computational machine learning technique SVM (Support Vector Machine) and pattern matching technique RIPOC (Rotation Invariant Phase Only Correlation). Object recognition and pose estimation are performed by SVM and their positions are estimated by RIPOC. A system using the proposed method can recognize objects and estimate their positions and poses from image sequences without special image marks or sensors. Experimental results of over 12, 000 images show high performance for object recognition and their position-pose estimation.
  • 堤 富士雄, 中島 慶人
    2001 年 121 巻 10 号 p. 1524-1529
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    Video database and object recognition have been treated as separate problems in the past. Recent retrieval applications achieved rapid indexing and robust retrieval capabilities, but the precision of recognizing objects in video images is lower than object recognition using machine learning. In contrast with video retrieval, machine learning needs vast computational training time in advance and cannot handle similarity easily. To solve these problems, we present an image-based video recognition framework combined with video retrieval and object recognition. To develop an effective combination, we evaluated several retrieval methods and the support vector machine(SVM), which is one of the most popular supervised learning techniques. From experimental results, we found that the combination of extended color-pair retrieval and SVM using color location is the most effective pair for high precision and rapid indexing of video recognition system.
  • 松本 圭介, 須藤 智, 斎藤 英雄, 小沢 慎治
    2001 年 121 巻 10 号 p. 1530-1539
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper, a new method for selecting a camera viewpoint automatically for soccer scene by tracking a ball is proposed. The soccer scene is taken by multiple fixed cameras. First, template matching is imple-mented for every image taken with each camera to detect and to track the ball. If the ball is occluded in one of images, the ball is tracked by using the position of the extracted ball in other camera. The tracked ball positions are integrated by using the geometrical relationship between cameras that is represented by Fundamental Matrix. For selecting the camera view, we define an evaluation function for camera selection based on tracking information of the ball. The system selects camera viewpoint of each frame automatically by finding the camera that gives maximum evaluation of camera selection. Experimental results demonstrate that the camera viewpoint is successfully selected for soccer scene by using the proposed system.
  • 濱本 和彦
    2001 年 121 巻 10 号 p. 1540-1548
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    The purpose of this paper is to develop compression method for medical ultrasonic echo image which is able to keep sufficient image quality for diagnosis. There are two ideas for it. One is a method to reduce the data volume without perceived loss of image quality in all region. The other is a method in which image quality in ROI is scarcely lowered and in the other area is severely lowered. In image compression with a JPEG baseline system, it is possible to control the compression ratio and image quality by controlling quantization values. The JPEG baseline system enables a JPEG bit stream to include four JPEG quantization tables. In this proposal, I divide an image into four regions which have different characteristic each other and attempt to optimize JPEG quantization table for each region. This method is named “Extended JPEG system” Medical ultrasoinc echo images are compressed by this method on ideas mentioned above. The compressed images are evaluated by rate, PSNR and the subjective assessment of the quality of pictures.
  • 宮腰 隆, 畠山 豊正, 澤柳 一美, 松田 秀雄, 中嶋 芳雄
    2001 年 121 巻 10 号 p. 1549-1554
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    Based on the experimentally determined infection process, Movable Finite Automaton (MFA) models of bacteriophage (phage) T4 are proposed, which describe the process of adsorption and injection of DNA into E. coli, starting from the initial state of floating.
    An MFA model for the phage T4 is constructed with eight parts of the phage structure. Using the model system, the infection process of phage T4 is simulated including six discrete stages; i.e., floating, landing on the host bacterium, E. coli, searching for receptor LPS, structural change of the baseplate, contraction of the tail sheath and injection of DNA. Each MFA model of eight parts has a state function(Qn) and some transition functions(fQn) based on biophysical considerations. Qn has a number of states that describe the chage of action in each part. In the execution of 3D simulation which utilized OpenGL, the actions of each stages can be reasonably well reproduced. MFA models for phage T4 and the phage actions at each stage are described briefly, and the results and problems of this model system are discussed.
  • 斉藤 文彦
    2001 年 121 巻 10 号 p. 1555-1561
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    The characters written on a signboard in a scene image are important information to develop a feature traffic support system and to search the objective image in a large image database. This paper proposes a method to extract the character string on a signboard in a gray-scaled scene image that is taken on the street. The location, size, inclination, vertical or horizontal direction, the number of characters and the type of character font are not fixed in a character string region in a scene image. In the method, the character string region is extracted using the genetic algorithm that evaluates the averaged strength of edges, the size of the region, the distribution of edge directions and the distribution of gray levels surrounding the region. The experimental results show that the character string regions were extracted with reliable recognition rate by the proposed method.
  • Naonobu Okazaki, Yoshimasa Baba, Tetsuo Ideguchi, Ken-ichi Nakata, Mi ...
    2001 年 121 巻 10 号 p. 1562-1569
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    Constructions of networks with the private IP addresses, i.e., private networks, become very popular. In general, private addresses may overlap each other. When we manage network equipment in private networks by a remote network manager using network management protocols, each private network requires its own network manager because the manager can not be shared by overlapping private networks. This leads to problems of cost and inefficiency. In this paper, we discuss an efficient method for network management of private networks using the concept of IP address translations. In this method, a new mechanism will be proposed, in which, the globally unique address mapping of SNMP data in address translation will be guaranteed. We will also discuss a trial implementation of the proposed method. By evaluation of network management system models, we will show that the proposed method is effective to reduce the cost and space of equipment in case of a large system including many private networks.
  • 村田 充利, 北村 敏明, 下代 雅啓, 山田 徹, 沢 新之輔
    2001 年 121 巻 10 号 p. 1570-1576
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper we propose a new type of laminated planar filters consisting of two quarter-wavelength mi-crostrip resonators located on both sides of the ground plane having a coupling slot. It is shown through numerical simulations by mean of the FD-TD method that the frequency characteristics of both pass band and attenuation pole are controllable by choosing the structural parameters appropriately.
  • 海老根 秀之, 志賀 泰博, 池田 美鈴, 中村 納
    2001 年 121 巻 10 号 p. 1577-1585
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    Several researchers have recently attempted automatic recognition and tracking of the facial expressions in video frames. However, it is very hard to detect a neutral face because the distinction between facial movement and head movement is difficult to make. To solve this distinction problem, the head position was initialized using the central point of each eye and the tip of the nose. The transition of facial expressions can be recognized by means of detecting the reference face (neutral face) and the recognition algorithm constructed to the expert system.
  • 須田 宇宙, 新谷 泰男, 浮貝 雅裕, 三井田 惇郎
    2001 年 121 巻 10 号 p. 1586-1592
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    This paper presents a new method to extract moving objects from image sequence both during and after the stop of their movement. It utilizes three frame memories, in which input signals based on a frame rate are stored in a pile with a certain ratio, respectively. A novel feature of this method is that it has introduced time variable parameters for accumulation of signals which extends the former theory. Based on simulation results, moving objects can be extracted and would not disappear for a certain time even after the stop of their movement. In addition, the effectiveness of time variable parameters is recognized to keep extracted objects longer compared to the former theory. From the results of some experiments, it is also confirmed that the method proposed in this paper is useful for moving objects extraction in the field of image processing.
  • 三浦 直樹, 渡部 彰一郎, 志田 和人, 福田 寛, 川添 良幸, 清水 俊夫
    2001 年 121 巻 10 号 p. 1593-1598
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    The generation of standard brain image is one of the most important subjects in brain image analysis. In this paper, we have developed a method that generates a standard brain image of transversal sectional view from a dataset of MRI (magnetic resonance imaging) brain images. First, every pair of 2 MR images in the dataset is registered by using a genetic algorithm, and the shape difference between 2 images is parameterized in this process to select the most representative standard image. Next, we generate a standard brain image by taking an average among the images that are registered to the representative standard image. It is expected that the 3D standard brain image would be also obtained similarly by extending this method.
  • 淺間 正義, 川畑 健太郎, 曽禰 元隆, 鵜飼 裕之
    2001 年 121 巻 10 号 p. 1599-1605
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    This paper proposes the digital excitation control system based on the H control for an engine generator. After the mathematical model for control design is derived from the equivalent circuit of a generator, the problem is formulated as the mixed sensitivity problem. Modeling errors and measurement noise to design weight functions are identified by step response experiments both without load and with maximum rate load. The effectiveness of proposed method is verified by both some numerical simulations and experiments using 4-poles 5kVA engine generator. As a result the proposed system represents robust control performance compared with the classical PID controller designed Ziegler-Nichols's method. Since the obtained H controller becomes high dimensional system, it usually takes much calculation time. From the view point of practical implementation to realize fast calculation, the digital AVR based on DSP (Digital Signal Processor) is used. Implementing the controller, the digital PWM pulse generation arlgorithm is applied.
  • 外乱分離からの考察
    阿部 直人
    2001 年 121 巻 10 号 p. 1606-1611
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    A limiting optimal regulator problems of disturbance decoupling for systems with input time delay are discussed. The limiting optimal regulator is derived by the weighting matrix with state vectors tend to infinity in a linear quadratic cost function. In the finite dimensional case, the limiting optimal regulator is known as a cheap control, which has a perfect regulation property and disturbance decoupling. Time delay systems are regarded as non-minimum phase systems. In this paper, however, it is shown that the limiting state feedback control achieves disturbance decoupling after the time-delay period.
  • 山口 智, 板倉 秀清
    2001 年 121 巻 10 号 p. 1612-1619
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    A new learning algorithm for agents to move around and to accomplish a task in unknown environments, is proposed. The actions of agents are a kind of the stochastic phenomena, so that, as an evaluation function of behavior of the agent, the likelihood of probabilities for reaching desired and undesired states, is utilized in our learning algorithm. In order to maximize the likelihood, two numerical computation methods are tried to be utilized: the dynamic programming and the hill climbing method.
    The proposed algorithms are applied two kinds of simulation to be compared with reinforcement learning algorithms. Through the computer simulations, the difference between our algorithm and the reinforcement learning algorithm is clarified.
  • 三上 弾, 萩原 将文
    2001 年 121 巻 10 号 p. 1620-1626
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper, we propose a self-growing learning vector quantization with additional learning(SGLVQ). It has high pattern recognition ability. In addition, additional learning is possible. The proposed SGLVQ is constructed based on self-organizing map (SOM) and learning vector quantization (LVQ) and is composed of 2 layers: the input/output layer and the rule layer. There are full connections between the input/output layer and the rule layer. Learning of the SGLVQ consists of 2 steps: SOM step and LVQ step.In the SOM step, similar inputs are mapped in the neighboring area using SUM algorithm. In the LVQ step, neurons are incremented and size of the network is adjusted automatically. Neurons increased when mis-recognition oc-curs.The increment of neurons enables additional learning and contributes to obtain high recognition ability.
  • 中野 敏治, 田 雅杰, 三宮 信夫
    2001 年 121 巻 10 号 p. 1627-1633
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    A local search method is proposed for solving job shop scheduling problems in the case where the total number of search points is limited. The objective of the problem is to minimize the makespan. The solution is represented by an operation sequence. Generation of better sequence is done by a combined effort of the neighborhood search and the global search. The neighborhood search is performed by exchanging an operation (called negotiator) with its neighboring one in the operation sequence. The global search is done by selecting the negotiator at random from the operation seqence. The Gantt Chart and the corresponding objective value are obtained from the operation seqence by applying a mixed procedure of the semiactive scheduling method and the active scheduling method. The proposed method includes three parameters: the size of neighborhood.. the rate of two scheduling methods and the total number of search points. Their rec-ommendable values are given from the resultt of computation experiments. It is also shown in experimental results that the proposed method outperforms the genetic algolithm for large-scale problems in the case of limited number of search points.
  • 瀧田 航一朗, 長名 優子, 萩原 将文
    2001 年 121 巻 10 号 p. 1634-1640
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    In this paper, we propose a new hierarchical pulse neural network and its reinforcement learning algorithm with network extension. The proposed pulse neural network has three layers: an input layer, a hidden layer and an output layer. All of the neurons are pulse neurons. This network learns relations between input pulse sequences and the desired outputs by updating connection weights and adding neurons corresponding to the input-output relation dynamically. We carried out the computer simulation (computer tennis game) to confirm the performance of the proposed algorithm.
  • 永田 武, 大野 益見, 佐々木 博司, 藤田 秀紀
    2001 年 121 巻 10 号 p. 1641-1646
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    This paper proposes a multi-agent system to power system unit commitment problems. Multiagent is a new paradigm for developing software applications. Coordinating the behavior of autonomous agents is a key issue in agent oriented programming techniques today. Recently, agents are being used in a increasingly wide variety of applications, ranging from comparatively smal systems such as E-mail filters to large. open. complex systems such as air traffic control. Though some agent frameworks have been proposed in the power system field, the number of studies is limited.
    In this paper, we developed a power system unit commitment application by multi-agent architecture. Our multi-agent system has the following characteristics: (1) The system consists of a single facilitator agent, two mobile agents, and one or more generator agents which are elements of power system network. (2) The facilitator agent is developed to act as a manager for the process by using the singleton design pattern. The mobile agents migrate to generator-agents to increase or decrease their power generations. The generator agents have their operational dia. (3) Message object is developed tocommunicate between the agents using KQML-like object. The proposed approach is applied to a simple model system, and the results show that the multi-agent system is an efficient decentralized approach for solving power system unit commitment problems.
  • 〓 暁英, 福見 稔, 北 研二
    2001 年 121 巻 10 号 p. 1647-1653
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    This paper presents a new approach to improve retrieval performance of the vector space model (VSM) by utilizing user-supplied information of those documents that are relevant to the query. In addition to user's relevance feedback information, information such as original document similarities is incorporated into the retrieval model, which is built by using a sequence of linear transformations. High-dimensional and sparse vectors are then reduced by SVD (Singular Value Decomposition) and transformed into the low-dimensional vector space, namely the space representing the latent semantic meanings of words. A prototype system based on proposed model has been built, and the experiments have been carry out on two test collections, Medline collection and Cranfield collection. The rises of retrieval precision compared with LSI (Latent Se-mantic Indexing) model are 4.03%_??_22.49% and 20.99%_??_24.87% for two training data sets, and 0.01%_??_0.62% and 4.25%_??_5.48% for test data sets, respectively. The proposed method provides an approach that enables us to preserve user-supplied relevance information for a long term in the system and to use it later.
  • Xiaoqiu WANG, Hua LIN, Jianming LU, Takashi YAHAGI
    2001 年 121 巻 10 号 p. 1654-1655
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    There is currently a real need for fast reliable computational algorithms to improve the accuracy of the measurement of the periodic signal. A small bias of the periodic signal frequency will cause the consequent undesired effect of harmonic interference due to spectral leakage. In this letter, a novel adaptive sampling algorithm for periodic signal measurement is proposed. According to the DFT analysis, we know that the spectral leakage will be reduced greatly by using the proposed algorithm. Simulation results show the effec-tiveness of the proposed method.
  • 鶴沢 偉伸, 武田 和大, 中村 幸仁, 中山 茂
    2001 年 121 巻 10 号 p. 1656-1657
    発行日: 2001/10/01
    公開日: 2008/12/19
    ジャーナル フリー
    Virtual touch is one of Virtual Reality that reproduces a human sense on a computer. A sense of touch can be produced imitatively by using a force-feedback which transmits resistive force to the device for the manipulation. In this letter, we have developed a mouse device with the function of virtual touch in which double motors were installed to produce the force-feedback. We have tested the mouse device with virtual touch for two applications of a virtual spring and a file viewer with drug & drop operation.
feedback
Top