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
124 巻, 3 号
選択された号の論文の45件中1~45を表示しています
特集:実環境におけるマシンビジョン適用技術
特集解説
特集論文
  • 林 宏樹, 斉藤 文彦
    2004 年 124 巻 3 号 p. 606-612
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    The template matching is a method to search the target image area that is similar to the template image. The template matching is a simple method. But, the target image must have the right position as the template image generally. Therefore it requires large computational cost when the location and the inclination of a target image area are unknown. This paper proposes a method to search the target image area that has a free location and a free inclination by the separated template image. In this method, the template image is divided into block areas with 3 by 3 pixels. The image pattern in each block is rotated by 45 degrees and matching process is performed between rotated blocks and an objective image. The experimental results show that the proposed method had a higher reliability for matching in comparison with the conventional methods.
  • 伊藤 誠也, 金子 俊一
    2004 年 124 巻 3 号 p. 613-620
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    An efficient algorithm for calculating Selective Correlation Coefficient (SCC) is proposed, which is expected as stable and robust for ill-conditions in imaging: illumination change, noise, and occlusion. The SCC is calculated through a masking operation based on Increment Sign Correlation which is efficient and robust for the ill- conditions. An SSDA algorithm and initial thresholding are introduced to remove redundant computation for obtaining much more efficiency. Through many experiments with real images, from twice to 150 times efficiency in comparison with CC and the original SCC could be obtained for ill-conditioned search problems.
  • 広井 高志, 田中 麻紀, 宍戸 千絵, 渡辺 正浩
    2004 年 124 巻 3 号 p. 621-628
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    This paper reports an image-processing algorithm for robust inspection of LSI wafer patterns using SEM. In order to detect defects in a regular LSI pattern, a pair of long patterns is compared, blocked images are aligned, and defects are judged using the aligned images. The LSI wafer pattern is defined to consist of blank space, fine repetitive patterns, and unique patterns. Distortion of the SEM image is larger than the repetitive pattern pitch, requiring the system to keep track of the alignment in areas without pattern and to mitigate the indeterminacy of repetitive patterns. To satisfy these requirements, we suggest algorithm focused on consistency among local registration. Registration candidates in each block are calculated and a chain of the correct registration sequence, using candidate information in all the related blocks. To increase robustness, combination of two evaluation functions is proposed. Two-dimensional correlation and low pass filtered one-dimensional correlation are used. The latter is complement of the former from the perspective of finding fine repetitive pattern edge to ease finding correct route. Experimental evaluations confirm that most pattern cases can be inspected correctly using the proposed SEM inspection system.
  • 奥田 晴久, 橋本 学, 鷲見 和彦, 佐々木 和則
    2004 年 124 巻 3 号 p. 629-636
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In this paper, we propose the new template matching method with high speed, high accuracy and robustness, which works robustly under bad-conditioned situations as occlusion and shading. In this method, firstly, the Hierarchical Distributed Template Matching (HDTM) which we have already proposed makes rough positioning and pose estimation with high speed. Then, we estimate the optimum position and pose robustly by using the LMedS robust estimation and partial templates method. The experimental results for full-rotated target objects show that it achieved the accuracy of errors less than ±0.1 pixels and ±0.1 degrees, and the computing time were shorten up to about 1/20 to the conventional hierarchical search method.
  • 武口 智行, 金子 俊一
    2004 年 124 巻 3 号 p. 637-646
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    A novel approach of object recognition is proposed, by which multiple three dimensional objects of range contours with occlusion and/or observation defects are efficiently and robustly recognized in cluttered scene. In our approach, in order to make the object recognition insensitive to peculiar features, such as edges and surface color, a uniform voxel framing has been utilized to define local coordination and to generate local depth structures on the objects, which are called depth aspect images DAI. In this paper, some techniques for efficient object recognition using DAI matching are designed. The first one is hashing of DAI images based on two dimensional hash table by use of local geometric key features, which enables fast search in the DAI database. The second one in recognition is random sampling for the voxel coordination, followed by parameter design with probabilistic modeling of sampling. By use of those two techniques, many redundant candidates for objects can be prevented from matching computation, realizing efficient and robust recognition of multiple objects in the complicated scene. The third one is introduction of support spaces for DAI calculation whichcooperates with a fast statistical evaluation of the similarity. We demonstrate recognition examinations to clutter scenes with a large DAI database involving eight models and verify the effectiveness and efficiency of these techniques through analysis of recognition on 100 scenes.
  • 田村 仁, 酒井 憲司
    2004 年 124 巻 3 号 p. 647-652
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    The method of measuring amount of piled-up objects using texture images analysis is developed and examined. This method treats an image of piled-up objects as a random texture. 9 type×5 size of minute shape features are extracted from the texture, and effective features to measure are statistically selected to derive a regression. When experimenting on 50-500 rubber bands in a box of 21.5×13.5cm, a regression that confidence interval of 99% is ±13.6 was obtained. However, because the feature of textures did not change any more when there were more rubber bands, it was not possible to estimate. As a result of another experiment that estimate dry mass weight of weeds in test grids of 70×70cm of the field, a regression that confidence interval of 99% is ±123.1g was obtained. The saturation problem is not caused in the estimation because there is a limit in the amount of the weed that grows in the test grid. The measurement accuracy is improved by distinguishing a predominant species in the grid as a preprocessing and selecting a suitable regression to the species thereafter.
  • Young Jun Roh, Hyoungsuck Cho
    2004 年 124 巻 3 号 p. 653-658
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In an x-ray vision, identifying the pose, i. e. the position and orientation of objects from x-ray projection images is extremely important to monitor in real time and analyze mechanical parts which are invisible from outside. It is assumed here that the x-ray imaging conditions that include the relative coordinates of the x-ray source and the image plane are predetermined and the object geometry is known. In this situation, an x-ray image of an object at a given pose can be estimated computationally by using a priori known x-ray projection image model. It is based on the assumption that a pose of an object can be determined uniquely to a given x-ray projection image. Thus, once we have the numerical model of x-ray imaging process, x-ray image of the known object at any pose could be estimated. Then, among these estimated images, the best matched image could be searched and found. When adequate features in the images are available instead of the image itself, the problem becomes easier and simpler. In this work, we propose an efficient pose estimation algorithm for polyhedral objects whose image features consist of corner points and edge lines in their projection images. Based on the corner points and lines found in the images, the best-matched pose of a polyhedral object can be determined. To achieve this, we proposed an adequate and efficient image processing algorithm to extract the features of objects in x-ray images. The performance of the algorithm is discussed in detail including the limitations of the method. To evaluate the performance of the proposed method a series of simulation studies is carried out for various imaging conditions.
  • 久保田 整, 小野 勝一, 武士 正美, 斎藤 英雄
    2004 年 124 巻 3 号 p. 659-665
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In this paper, we propose a method to estimate 3D shape of deformable plastic tapes from multiple camera images. In this method, the tape is modeled as serial connection of multiple rectangular plates, where the size of each plate is previously known and node angles of between plates represent the shape of the object. The node angles of the object are estimated by 2D silhouette shapes taken in the multiple images. The estimation is performed by minimizing the difference of the silhouette shapes between the input images and synthesized images of the model shape. For demonstrating the proposed method, 3D shape of a tape is estimated with two camera images. The accuracy of the estimation is sufficient for making the assembling robot in our plant to handle the tape. Computation time is also sufficiently short for applying the proposed algorithm in the assembling plant.
  • 柏木 利幸, 大恵 俊一郎
    2004 年 124 巻 3 号 p. 666-671
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    We propose a new detection method of non-uniform color area. This method discriminates non-uniform area from uniform area by color frequency. This frequency is obtained by using color histogram, in which each point has the frequency of the color. The frequency of the color of non-uniform area is low compared with the uniform area because the non-uniform area is smaller than the uniform area. To compare the frequency, we create gray histogram images in which original color value is replaced by the frequency of the color.
    This method requires very few amounts of calculation although it needs multiple color images. Moreover, this method gives us new evaluation criteria for color uniformity.
  • Xin Lu, Shunichiro Oe
    2004 年 124 巻 3 号 p. 672-679
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    This paper presents a color tracking method based on probabilistic data association in order to resolve difficult and complicated visual tracking problem, such as a changing of target’s representation, a clutter of environments and an interaction of target and camera. Because the probabilistic data association is flexible and suitable for ambiguous and missing data, which generates the difficulties of visual tracking, some methods of probabilistic data association could be combined and applied in this tracking method to find the solutions of these difficulties. Due to using sequential Monte Carlo framework, this tracking method is applied to tracking of changeful target by handling the related information between every frame in image sequences. In order to improve tracking accuracy, this method utilizes factorized sampling algorithm to express target characters as sample-set. Moreover, this method benefits from HSV color model and captures the color natures of object like human to enhance the color-sensing capability of computer. Hence, this method could be considered as self-learning system and imitate the based human vision function - tracking. The tracking system applying this method is implemented in real-time at around 15Hz with 640 × 480 pixels image. The results show that the self-learning and real-time system is able to track a target robustly with enough accuracy and automatically control the camera’s pan, tilt and zoom to remain the object centered in the field of vision.
  • 渡邊 睦, 西 奈津子
    2004 年 124 巻 3 号 p. 680-688
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    The authors are developing a vision-based intension transfer technique by recognizing user’s face expressions and movements, to help free and convenient communications with aged or disabled persons who find difficulties in talking, discriminating small character prints and operating keyboards by hands and fingers. In this paper we report a prototype system, where layered daily conversations are successively selected by recognizing the transition in shape of user’s mouth parts using camera image sequences settled in front of the user.
    Four mouth part patterns are used in the system. A method that automatically recognizes these patterns by analyzing the intensity histogram data around the mouth region is newly developed. The confirmation of a selection on the way is executed by detecting the open and shut movements of mouth through the temporal change in intensity histogram data.
    The method has been installed in a desktop PC by VC++ programs. Experimental results of mouth shape pattern recognition by twenty-five persons have shown the effectiveness of the method.
  • 坂田 幸辰, 金子 俊一, 高木 裕治, 奥田 浩人
    2004 年 124 巻 3 号 p. 689-698
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    A novel sequential learning algorithm of Test Feature Classifier (TFC) which is non-parametric and effective even for small data is proposed for efficiently handling consecutively provided training data. Fundamental characteristics of the sequential learning are examined. In the learning, after recognition of a set of unknown objects, they are fed into the classifier in order to obtain a modified classifier. We propose an efficient algorithm for reconstruction of prime tests, which are irreducible combinations of features which are capable to discriminate training patterns into correct classes, is formalized in cases of addition and removal of training patterns. Some strategies for the modification of training patterns are investigated with respect to their precision and performance by use of real pattern data. A real world problem of classification of defects on wafer images has been tackled by the proposed classifier, obtaining excellent performance even through efficient modification strategies.
  • 小川 行宏, 加藤 邦人, 山本 和彦
    2004 年 124 巻 3 号 p. 699-705
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    We describe realization of more complicated function by using the information acquired from some equipped unripe functions. The self-learning and recognition system of the human facial expression, which achieved under the natural relation between human and robot, are proposed. The robot with this system can understand human facial expressions and behave according to their facial expressions after the completion of learning process. The system modelled after the process that a baby learns his/her parents’ facial expressions. Equipping the robot with a camera the system can get face images and equipping the CdS sensors on the robot’s head the robot can get the information of human action. Using the information of these sensors, the robot can get feature of each facial expression. After self-learning is completed, when a person changed his facial expression in front of the robot, the robot operates actions under the relevant facial expression.
  • 岩田 健司, 本郷 仁志, 山本 和彦, 丹羽 義典
    2004 年 124 巻 3 号 p. 706-712
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    Positions of facial parts, such as eyes, mouth, and nose, are required for constructing the user interface by face image recognition or for face pose estimation. We propose a facial parts detection method that is robust for face postures. The method uses four directional features and relaxation matching. Facial part templates based on four directional features learned from many face images correspond to varieties of input pattern. The initial probability is obtained by template matching of the four directional features. Positions of facial parts are detected by relaxation matching using the spring connection of each facial part. Since it corresponds to various face postures, the multi-direction integrated model is used. The proposal method is short computation time and useful for real-time system. Experiments show that facial parts are detectable to the HOIP multi-directional face image database. Detection rate was 96% for ±45°horizontal and ±30°vertical face postures.
  • 山口 順一, 清家 裕
    2004 年 124 巻 3 号 p. 713-719
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In face recognition (face verification, face expression etc.), a full face or near full face is used and the face image is about fixed size in general. Especially, eyes, nose and mouth are usually located from the upper part to the lower part in the input image. But, in order to recognize the face in any posture, it is important to remove an influence caused by the position and three-dimensional turning of the face. The authors propose a method for detecting the face position in unknown posture, using an invariant image information. First, we show that the spectrum, which is obtained by polar transform and Fourier transform of the image, is shift-invariant and rotation-invariant, and is shift-invariant toward depth. Next, we describe on the detection of the face position in unrestricted posture, using the calculation of correlation of the spectrum. In this paper, the proposal method is explained and the experimental result, which is performed to verify the efficacy of the method, is demonstrated.
  • 陳 彬, 目黒 光彦, 金子 正秀
    2004 年 124 巻 3 号 p. 720-728
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    This paper proposes a method to estimate sound source position by fusing auditory and visual information with Bayesian network in human-robot interaction. We firstly integrate multi-channel audio signals and a depth image about the environment to generate a likelihood map for sound source localization. However, this integration, denoted by “MICs", does not always lead to locate a sound source correctly. For correcting the failure in localization, we integrate the likelihood values generated from “MICs" and the skin-color distribution in an image according to the result of classifying audio signal into speech/non-speech categories. The audio classifier is based on support vector machine(SVM) and skin-color distribution is modeled by Gaussian mixture model (GMM). With the evidences given by MICs, SVM and GMM, we infer whether pixels in images correspond to a sound source or not according to the trained Bayesian network. Finally, experimental results are presented to show the effectiveness of the proposed method.
  • 西口 敏司, 東 和秀, 亀田 能成, 角所 考, 美濃 導彦
    2004 年 124 巻 3 号 p. 729-739
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    It is useful for automatic video shooting in a lecture room to estimate the location of a speaker in the lecture room. The captured videos are used for distance learning and lecture archiving systems. In order to estimate the location of a speaker in a wide lecture room, multiple cameras and multiple microphones are used. However, it is difficult to estimate the precise location of a speaker using only visual or acoustic sensors because of calibration problems, noise, and other interference. Therefore, we propose a method that integrates audio and visual information from a speaker in the lecture room. A lecturer’s cell and a student’s cell ared introduced as a unit of estimation of the location of a speaker. We defined 120 cells in a real lecture room and our multi-modal method were applied to the cells. The estimation accuracy of the location of a speaker is sufficient for automatic video shooting of a speaker in a lecture room by our integrating method.
  • 渡辺 隆, 輿水 大和
    2004 年 124 巻 3 号 p. 740-747
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    This paper proposes an image processing method to enforce the quality control of the molded electronic devices on the PC environment in the real production line. It was known that the cost of this inspection system could be expected below the real tact time and the specified running cost. Since the plastic mold of the electronic devices is likely to be distorted and to suffer from the defects called “bari"(in Japanese) which is a kind of tiny chip attached at the wedge of the mold, several degradations of the quality will be increased in the fabrication process. If such defects could not be detected in advance, for example, the fabrication of the electric terminals to the mold can not be successful at the end of the process. The proposed algorithm was implemented on the general purpose PC in the real production line and was proved to be practical by means of more than 6,000,000 device inspections in 6 month.
  • 高木 裕治, 渋谷 久恵, 細谷 直樹
    2004 年 124 巻 3 号 p. 748-752
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    Hi-speed and highly sensitive visual inspection tools are in practical use in the field of semiconductor wafer inspection. Defect review is a critical step to grasp the defect condition, that is, to know what types of defects are occurred on the wafer. However, it is not practical to review whole defects on the wafer because of a quantitative gap between the detected defects and defects to be reviewed within certain practical time. To bridge this gap, a visual pattern detection algorithm of spatially distributed defect points and an optimum defect sampling plan are proposed. Experimental result shows that partial defect review is enough to grasp the defect occurrence condition all over the wafer. by applying the proposed method for defect review.
  • 山田 宗男, 上田 浩次, 堀場 勇夫, 津川 定之, 山本 新
    2004 年 124 巻 3 号 p. 753-760
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    This paper relates to the study of the road surface condition detection technique based on the image taken by TV camera attached to the rearview mirror of a vehicle. We examine a technique to detect features related to water and snow on the road surface, in which the detection was stably possible throughout day and night. The extraction of the water on the road is made with the technique which gets the property of the polarization from the image. And the extraction of the snow on the road is made with the texture analysis. We conducted field tests to verify the detection ability of the road surface condition while running on the expressway at 100km/h on the average and obtained a favorable result.
  • 竹内 英世, 松本 憲幸, 桑山 清明, 梅崎 太造
    2004 年 124 巻 3 号 p. 761-768
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In this paper, we propose the way to evaluate fingerprint image-quality and how to discriminate remnants from captured images.
    First, we investigate evaluating fingerprint image-quality. Fingerprint image-quality can be digitized using the "measure" we proposed. We simulate using the dataset consists of 1425 fingerprint images captured from 57 people in Feb, which contains a lot of faded images. In the simulation using all our database, recognition rate is 95.6% while type II error is 0.01%. Recognition rate is improved to 98.1%, with rejecting 3.7% faded images evaluated by our measure from the database. Recognition rate is improved to 99.6%, rejecting 14.2% faded images. And we investigate the way to apply the measure of image-quality to fingerprint verification device with customer’s satisfaction in real world.
    Next we propose the way to discriminate between remnants and fingerprint images captured from optical scanner by using frequency analysis. We can perfectly prevent the fingerprint verification device from malfunctioning caused by remnant, when strong flashlight or direct sunlight slant in optical scanner in real world.
  • 川野 卓也, 山本 和彦, 加藤 邦人, 本郷 仁志
    2004 年 124 巻 3 号 p. 769-775
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In this paper, we propose an electric appliance control support system for aged and bedridden people using pose recognition. We proposed a pose recognition system that distinguishes between seven poses of the user on the bed. First, the face and arm regions of the user are detected by using the skin color. Our system focuses a recognition region surrounding the face region. Next, the higher order local autocorrelation features within the region are extracted. The linear discriminant analysis creates the coefficient matrix that can optimally distinguish among training data from the seven poses. Our algorithm can recognize the seven poses even if the subject wears different clothes and slightly shifts or slants on the bed. From the experimental results, our system achieved an accuracy rate of over 99 %. Then, we show that it possibles to construct one of a user-friendly system.
  • 足立 和正, 濱田 尊裕, 中野 倫明, 山本 新
    2004 年 124 巻 3 号 p. 776-783
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    Drowsiness in car-driving relates to eye-blinking. Authors have developed the robust image capture CCD system by moving image processing for the brightness of day and night in a car, using the weak pulsed infrared LED’s light. This effect was evaluated by real circumstances in a car: darkness in night and brightness in direct sunshine. To extract upper and lower eyelids’ fringes from noises, a pair of edges from a pupil was used. The second derivative to blinking waveform was introduced to obtain the stable blinks, eliminating the different characteristics among individuals’ ones. The experiment to presume the drowsiness was referred to the steering wheel movement. This tentative experiment indicated good correlation between drowsiness and eye closure time.
  • 平山 高嗣, 岩井 儀雄, 谷内田 正彦
    2004 年 124 巻 3 号 p. 784-797
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    A security system using biometric person authentication technologies is suited to various high-security situations. The technology based on face recognition has advantages such as lower user’s resistance and lower stress. However, facial appearances change according to facial pose, expression, lighting, and age. We have developed the FACELOCK security system based on our face recognition methods. Our methods are robust for various facial appearances except facial pose. Our system consists of clients and a server. The client communicates with the server through our protocol over a LAN. Users of our system do not need to be careful about their facial appearance.
  • 谷 英哲, 田中 佳代, 東久保 政勝, 寺田 賢治, 山口 順一
    2004 年 124 巻 3 号 p. 798-804
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In this paper, we propose a method of detecting pedestrians on crosswalk by CCD camera. By detecting pedestrians, we can control the time of green light for the handicapped and the aged. First, measurement areas are defined along white line of crosswalk. Next, make space-time images that a slip of image on measurement area is arranged. Finally, pedestrians are detected by processing the image. In addition, we show the result of experiment detecting pedestrians in order to show that this method is effective.
  • 依田 育士, 細谷 大輔, 坂上 勝彦
    2004 年 124 巻 3 号 p. 805-811
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    Dozens of people are killed every year when they fall off of train platforms, making this an urgent issue to be addressed by the railroads, especially in the major cities. This concern prompted the present work that is now in progress to develop a Ubiquitous Stereo Vision based system for safety management at the edge of rail station platforms. In this approach, a series of stereo cameras are installed in a row on the ceiling that are pointed downward at the edge of the platform to monitor the disposition of people waiting for the train. The purpose of the system is to determine automatically and in real-time whether anyone or anything is in the danger zone at the very edge of the platform, whether anyone has actually fallen off the platform, or whether there is any sign of these things happening. The system could be configured to automatically switch over to a surveillance monitor or automatically connect to an emergency brake system in the event of trouble.
特集研究開発レター
論文
<電子デバイス>
<電子・集積回路>
  • 星野 聖彰, 荒木 英夫, 久津輪 敏郎, 原嶋 勝美
    2004 年 124 巻 3 号 p. 819-826
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    This paper proposes a new logical verification method that generates a test bench automatically from a software program. It is suitable to develop a System-On-a-Chip (SoC) device by co-designing of hardware and software. In designing an LSI device designer have to supply its test bench and its input signal patterns for the logical verification manually, because there is no formalized technique of generating them automatically. In such situations, the high degree integration of a device realizes the SoC technology that implements processors, memories, etc. on one device. In order to solve the complicated description of test benches and input signal patterns required for the logical verification, we propose a description technique, called ALET language (an Assembly Language which makes it Easy to generate the Test bench), with which the designer can enter the hardware test items into the assembly language file of its software. In result, the more the test item increases, the more the description decreases. In the case of 100 test items, the amount of required inputs is reduced to approximately 6% of the former. Consequently, the testing becomes easier by using our proposing logical verification technique corresponding to high integration of a device.
<通信・ネットワーク>
<マルチメディア>
  • Xugang Wang, Zheng Tang, Hiroki Tamura, Masahiro Ishii
    2004 年 124 巻 3 号 p. 842-851
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    A new multi-layer artificial neural network learning algorithm based on pattern search method is proposed. The learning model has two phases-a pattern search phase, and a local minimum-escaping phase. In the pattern search phase, our method performs local search iteratively and minimize the error measure function along with the set of descent directions of the error measure directly and finds the nearest minima efficiently. When the network gets stuck in local minima, the local minimum-escaping phase attempts to fill up the valley by modifying temperature parameters in ascent direction of the error measure. Thus, the two phases are repeated until the network gets out of local minima. The learning model is designed to provide a very simple and effective means of searching the minima of objective function directly without any knowledge of its derivatives. We test this algorithm on benchmark problems, such as exclusive-or (XOR), parity, Arabic numerals recognition, function approximation problems and a real world classification task. For all problems, the systems are shown be trained efficiently by our method. As a simple direct search method, it can be applied in hardware implementations easily.
<音声画像処理・認識>
  • 祖 国威, 大山 航, 若林 哲史, 木村 文隆
    2004 年 124 巻 3 号 p. 852-860
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    The basic process of automatic text classification is learning a classification scheme from training examples then using it to classify unseen textual documents. It is essentially the same as graphic or character pattern recognition process. So the pattern recognition approaches can be used for automatic text categorization. In this research several statistical classification techniques each of which employs Euclidean distance, various similarity measures, linear discriminant function, projection distance, modified projection distance, SVM, nearest-neighbor, have been used for automatic text classification. The principal component analysis was used to reduce the dimensionality of the feature vector. Comparative experiments have been conducted on the Reuters-21578 test collection of English newswire articles. The results illustrate that the efficiency of modified projection distance is totally better than the other methods and the principal component analysis is suitable for reducing the dimensionality of the text features.
<制御・計測>
<ソフトコンピューティング>
  • 金久保 正明, 萩原 将文
    2004 年 124 巻 3 号 p. 883-889
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In this paper, we propose an improved genetic algorithm based on the combination of Bee system and Inverse-elitism, both are effective strategies for the improvement of GA. In the Bee system, in the beginning, each chromosome tries to find good solution individually as global search. When some chromosome is regarded as superior one, the other chromosomes try to find solution around there. However, since chromosomes for global search are generated randomly, Bee system lacks global search ability. On the other hand, in the Inverse-elitism, an inverse-elite whose gene values are reversed from the corresponding elite is produced. This strategy greatly contributes to diversification of chromosomes, but it lacks local search ability. In the proposed method, the Inverse-elitism with Pseudo-simplex method is employed for global search of Bee system in order to strengthen global search ability. In addition, it also has strong local search ability. The proposed method has synergistic effects of the three strategies. We confirmed validity and superior performance of the proposed method by computer simulations.
  • 田中 一男, 松永 和之, 堀 滋樹
    2004 年 124 巻 3 号 p. 890-896
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    This paper presents electroencephalogram-based control of a mobile robot. The control purpose is to achieve direction control of a mobile robot only by electroencephalogram. We develop an algorithm for detecting direction thinking (‘going left’ or ‘going right’) and apply it to direction control of a mobile robot. The detecting algorithm is based on time-frequency domain analysis using continuous wavelet transformation. Our experimental results demonstrate the possibility of achieving direction control of a mobile robot only by electroencephalogram.
  • 伊藤 一成, 高倉 康弘, 斎藤 博昭
    2004 年 124 巻 3 号 p. 897-903
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    In past studies, neurons in chaotic neural networks have sigmoid function as an activated function. This paper proposes a new chaotic neural networks using a non-monotonic activated function. This network generate chaotic dynamics by transforming a shape of the function. We apply this network to the memory search systems. In that case, we introduce an original control term to the networks. It leads neurons to give stronger signal outputs with characteristic condition agreements, and weaker signal for disagreements. Adding constraints to the state transitions of the network, the output of network becomes more changeable to the state where the condition is satisfied. Due to its effect, recalling on a target pattern in fewer steps is achieved on average. Performance of the memory search system has been also greatly improved with the number of memory patterns where the conventional methods with sigmoid functions hardly recalled. Furthermore, our memory search system shows a great improvement in the case that each stored pattern has high degree of correlation.
<システム>
<ソフトウェア・情報処理>
  • 大谷 哲夫
    2004 年 124 巻 3 号 p. 921-928
    発行日: 2004年
    公開日: 2004/06/01
    ジャーナル フリー
    We are developing an information and communication system for a demand area power system (DAPS), one of the next-generation of distribution grids to which many dispersed generators can connect freely. This information and communication system is required to be adaptive and autonomous because states of DAPS may change dynamically. In addition, it must perform quickly enough to isolate a fault section from other sections in DAPS. Mobile agent technique promises to make a system adaptive and autonomous. However, existing platforms for mobile agents do not take it into account to perform in real-time such as isolation of a fault section. In this paper, we propose a mobile agent platform named MAFDAPS (Mobile Agents For Demand Area Power Systems) that provides priority controls and object classes making itself flexible. We have built a prototype system of MAFDAPS using JavaTM. The results from this prototype system indicate the applicability of MAFDAPS to operations of DAPS.
研究開発レター
<通信・ネットワーク>
feedback
Top