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
127 巻, 4 号
選択された号の論文の28件中1~28を表示しています
特集:実利用マシンビジョン技術
特集解説
特集論文
<センシング>
  • 佐野 恵美子, 前田 卓志, 松下 雅仁, 鹿井 正博, 笹川 耕一, 近江 雅人, 春名 正光
    2007 年 127 巻 4 号 p. 472-479
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    Fingerprint has been widely used as the method to realize personal authentication because it has the best balance among authentication performance, cost, size of device, and ease of use. However, conventional fingerprint sensors have some problems; quality of the fingerprint pattern is sometimes not so good due to the condition of a finger surface such as moisture or wrinkles. These problems are caused by the principle of the sensors. Most of fingerprint sensors detect a fingerprint pattern from the presence/absence of the convexities and concavities of a finger surface. To solve these problems, we devised a new kind of fingerprint sensor, which detects a fingerprint pattern through the use of optical characteristics inside a finger. This method is based on a new scientific discovery that there is layer of skin inside a finger that has a transmittance distribution corresponding to the pattern of concavities and convexities of a finger surface. Therefore, it can detect stable fingerprint images at all times regardless of finger surface conditions (wet, wrinkles). In this paper, we describe the principle of a new type of fingerprint sensor and demonstrate the effectiveness of the proposed sensor by using some examples.
<音声画像処理・認識>
  • 山下 淳, 蔵本 昌之, 金子 透
    2007 年 127 巻 4 号 p. 480-488
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    In this paper, we propose a new method that can remove view-disturbing noises from images taken with multiple cameras. In outdoor environment, it is often the case that scenes taken by cameras are hard to see because of adherent noises on the surface of the lens protecting glass. The proposed method analyses multiple camera images describing the same scene, and synthesizes an image in which adherent noises are eliminated. We show the effectiveness of the proposed method through experimental results.
  • 泉 哲也, 服部 哲郎, 北島 博之, 山崎 敏範
    2007 年 127 巻 4 号 p. 489-496
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    In order to obtain a low computational cost method for automatic handwritten characters recognition, this paper proposes a combined system of two rough classification methods based on features of a vector field: one is an autocorrelation matrix method, and another is a low frequency Fourier expansion method. In each method, the similarity is defined as a weighted sum of the squared values of the inner product between input pattern and the reference patterns that are normalized eigenvectors of KL (Karhunen-Loeve) expansion. This paper also describes a way of deciding the weight coefficients using Linear Regression Method, and shows the effectiveness of the proposed method by illustrating some experimentation results for 3036 categories of handwritten Japanese characters.
  • 竹内 英世, 保黒 政大, 松本 憲幸, 梅崎 太造
    2007 年 127 巻 4 号 p. 497-505
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    In this paper, we propose the way to modify gradient of fingerprint images. We modify the gradient using fingerprint area and contour, without extracting end points or bifurcation points called “minutia" from fingerprint images. It takes 3.2[ms] to modify fingerprint gradient by using a personal computer which operates at 1000[MHz] clock frequency. Our program code is so light weight that it is able to be embedded in fingerprint verification devices which operate at lower speeds. We evaluated our algorithm by using fingerprint images tilted respectively 0°, 5°, 10° and 15°. Without modifying gradient, we achieved 90.8[%] recognition rate by images tilted from 0° to 5°, while false acceptance rate is 0.01[%]. By using our gradient modification algorithm, recognition rate raised from 90.8[%] to 98.2[%]. On the other hand, 19.4[%] recognition rate was achieved by images tilted from 10° to 15° without gradient modification, and the rate raised to 89.2[%] with gradient modification. The users who don't get used to fingerprint devices tend to input his finger tilted on scanner. Thus, it is expected that the fingerprint verirification device will be improved to be more friendly for all peoples, by using our gradient modification algorithm.
  • 岡部 亜梨子, 安倍 満, 小沢 慎治
    2007 年 127 巻 4 号 p. 506-512
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    We propose an algorithm to detect human position from an in-store image for a flow line analysis proceeded in industrial customer researches. An image processing in in-store image involves difficulties such as occlusions and noise due to low resolution. In addition, the distribution of normal surveillance camera requires algorithm capable of single camera detection. Applying AbS method, which uses synthesized images based on 3D model of the store to analyze real images, the detection resolving the problems is accomplished. The experimental result shows the validity of the algorithm with respect to occlusion caused by human as well as static objects.
  • 塙 圭二, 濱田 安之, 松尾 陽介
    2007 年 127 巻 4 号 p. 513-520
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    This paper presents a method to detect the position of crop rows, ridges and borders in agricultural fields. The focus of this study is to realize an operating assist system of agricultural vehicles. A stereo vision set is mounted on a vehicle and the images obtained from the stereo cameras are processed to make distance images. Three dimensional models of crop rows, ridges and field borders are prepared. The distance images are analyzed using these models applying the pattern matching technique. Experiments with controlling the vehicle to run along the crop rows, ridges and borders were carried out successfully.
  • 渋谷 久恵, 高木 裕治, 中川 泰夫
    2007 年 127 巻 4 号 p. 521-527
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    A spatial pattern recognition algorithm is proposed for the effective sampling method to characterize defect occurrence of an entire wafer by reviewing a small number of defect samples. The algorithm classifies defects into clustered defects or regional defects or random defects based on analysis of defect densities and distances between neighboring points. Voronoi diagrams are applied to calculate the densities and the distances. Regional defects are classified into four major classes: rings, blobs, lines and arcs. Ring and blob patterns are detected by template matching techniques, while line and arc patterns are detected by utilizing their geometric properties. The Hough transform is used to detect line patterns and the detected pattern is verified by analyzing features. The proposed algorithm was evaluated using 916 sample wafers obtained in a real semiconductor process. 93.3% of the sample wafers with ring or blob patterns (250/268) were processed correctly while 3.9% of the sample wafers on which ring or blob patterns are detected (13/332) were false. 95.1% of the sample wafers with line patterns (117/123) were processed correctly while 18.7% of the sample wafers on which line patterns are detected (70/375) were false.
  • 山足 和彦, 藤原 孝幸, 輿水 大和
    2007 年 127 巻 4 号 p. 528-536
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    Image processing is generally constructed based on the gray information such as statistical gray histogram, gradient, texture, etc. However, a concrete technique that used this co-occurrence procession had been chiefly limited to the texture identification. Since the gray histogram is the most influential for extracting the global image properties, it has been always expected to be improved or extended. From this view point, it was notable that Dr. Kashiwagi proposed a new paradigm of feature extraction called Frequency Image (FI). As well as histogram, co-occurrence histogram or co-occurrence matrix is also such influential basic image feature especially for the texture feature analysis. In this paper, we introduce a new paradigm of texture feature image called “Co-occurrence Frequency Image (CFI)” as a complete extension of FI.
  • 中村 克行, 邵 肖偉, 趙 卉菁, 柴崎 亮介
    2007 年 127 巻 4 号 p. 537-545
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    In this paper the authors propose a method for recognizing non-stationary walking based on a gait analysis using multiple laser range scanners. The proposed method consists of the following procedures: (1) people tracking; (2) detection of gait features; (3) recognition of non-stationary walking. First, people tracking is performed by recognizing patterns in which the range data obtained near ankle rhythmically. Next, gait analysis is performed by the spatio-temporal clustering using Mean Shift algorithm. Finally, One Class Support Vector Machine (One Class SVM) is applied for learning and classifying a non-stationary walking. The experiment in a station concourse in Tokyo shows the overall accuracy of 98.4% by the proposed method.
  • 藤原 孝幸, 山足 和彦, 輿水 大和
    2007 年 127 巻 4 号 p. 546-552
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    We have proposed that the co-occurrence frequency image (CFI) based on the co-occurrence frequency histogram of the gray value of an image has a new potential to introduce a new scheme for image feature extraction. This paper proposes a couple of new enhancement filters to achieve shaping and smoothing of images. These filters are very similar to but quite different from those which have been used so far. Thus, we could show a possibility for introducing a new paradigm for basic image enhancement filters.
  • 寺田 賢治, 熱田 宏司
    2007 年 127 巻 4 号 p. 553-560
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    Recently, many brutal crimes have shocked us. On the other hand, we have seen a decline in the number of solved crimes. Therefore, the importance of security and self-defense has increased more and more. As an example of self-defense, many surveillance cameras are set up in the buildings, homes and offices. But even if we want to detect a suspicious person, we cannot check the surveillance videos immediately so that huge number of image sequences is stored in each video system. In this paper, we propose an automatic method of generating passer-by record images by using internet camera. In first step, the process of recognizing passer-by is carried out using an image sequence obtained from the internet camera. Our method classifies the subject region into each person by using the space-time image. In addition, we obtain the information of the time, direction and number of passey-by persons from this space-time image. Next, the present method detects five characteristics: the gravity of center, the position of person's head, the brightness, the size, and the shape of person. Finaly, an image of each person is selected among the image sequence by integrating five characteristics, and is added into the passer-by record image. Some experimental results using a simple experimental system are also reported, which indicate effectiveness of the proposed method. In most scenes, the every persons was able to be detected by the proposed method and the passer-by record image was generated.
  • 長 元気, 盧 存偉
    2007 年 127 巻 4 号 p. 561-567
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    Most of the recent 3-D image measurement methods have employed certain pattern projection techniques because of their higher reliabilities. Especially, the intensity-modulated technique can detect more stripe addresses by a single projection and therefore is very much expected for its practical use in near future. The traditional techniques of this type, however, have encountered several serious problems where 3-D measurements are difficult in sensitivity, speed and accuracy or fatally impossible for unknown objects. In order to solve the this problems, we propose an automatic control technique of angle and space frequency of projection pattern. Furthermore, for secure of measurement accuracy and measurement speed, projection pattern intensity control technique and optimal intensity-modulation projection technique are adopted in the proposal measurement system. By using proposed technique the automatic 3-D measurement intended for the geostationary objects and the swaying objects were realized in experimental results.
  • Wei Jiang, Jian Lu
    2007 年 127 巻 4 号 p. 568-576
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    Technology for capturing panoramic (360 degrees) three-dimensional information in a real environment have many applications in fields: virtual and complex reality, security, robot navigation, and so forth. In this study, we examine an acquisition device constructed of a regular CCD camera and a 2D laser range scanner, along with a technique for panoramic 3D reconstruction using a data fusion algorithm based on an energy minimization framework. The acquisition device can capture two types of data of a panoramic scene without occlusion between two sensors: a dense spatio-temporal volume from a camera and distance information from a laser scanner. We resample the dense spatio-temporal volume for generating a dense multi-perspective panorama that has equal spatial resolution to that of the original images acquired using a regular camera, and also estimate a dense panoramic depth-map corresponding to the generated reference panorama by extracting trajectories from the dense spatio-temporal volume with a selecting camera. Moreover, for determining distance information robustly, we propose a data fusion algorithm that is embedded into an energy minimization framework that incorporates active depth measurements using a 2D laser range scanner and passive geometry reconstruction from an image sequence obtained using the CCD camera. Thereby, measurement precision and robustness can be improved beyond those available by conventional methods using either passive geometry reconstruction (stereo vision) or a laser range scanner. Experimental results using both synthetic and actual images show that our approach can produce high-quality panoramas and perform accurate 3D reconstruction in a panoramic environment.
  • 田中 志歩, 伊東 敏夫, 大川 剛直
    2007 年 127 巻 4 号 p. 577-582
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    In this paper, we propose the matching method using the edge image with the rigid assumption. We introduce the virtual intensity added to the edge image and regard it as energy function. It is assumed that the objects to match each other is rigid. The edges are repeatedly moved to the direction which the energy function are reduced by parallel translation and rotation. The corresponding point between two images can be found when the function is minimum. Real images include the area which have no or extreme intensity change. The proposed method is available in those cases since it uses features of the image that is the edges. We showed this advantage by the experiments of motion estimation.
  • 鈴木 康弘, 山本 和彦, 加藤 邦人, 安藤 道則, 小島 真一
    2007 年 127 巻 4 号 p. 583-590
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    Many active safety technologies for the driver support system are developing. Most of the traffic accidents are caused by driver's inattentive or drowsy. We are developing a driver support system that protects from traffic accidents by these causes. Our purpose is to detect the driver's face region by using a camera. A lot of face detection methods are proposed, but there is not a technique addressing every environment inside the car. For example, skin color segmentation can not detect the skin region in the night, because it has to light up the driver by bright light. In this paper, we propose a skin detection method by the unique reflection characteristics of the materials. Our method is very simple algorithm. We developed a skin detection system, and confirmed effectiveness by the evaluation experiment in indoor environment, and showed the effectiveness by a driving experiment in the night.
  • 坂下 祐輔, 藤吉 弘亘, 平田 豊, 高丸 尚教, 深谷 直樹
    2007 年 127 巻 4 号 p. 591-598
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    This paper presents a method for real-time cycloduction measurement based on fast ellipse detection. The Hough transform frame work is known to be effective for ellipse fitting, but most of existing methods need a high computation cost. We therefore present a novel technique for detecting a region of the pupil by fitting an ellipse based on parallelogram. Since our method eliminates outliers due to lighting, it is possible to obtain the ellipse parameters in high accuracy. Once the ellipse parameter for pupil image is detected, we can generate a projected image of iris pattern, and then estimate a rotation angle of the cycloduction by SAD matching considering the temporal information. Our experimental results demonstrate the effectiveness and robustness of our method.
  • 関 真規人, 谷口 博康, 橋本 学
    2007 年 127 巻 4 号 p. 599-604
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    The overtopping wave which is high wave exceeding breakwaters causes damage to a coast area. In this paper, we describe a novel detection method of the overtopping wave and other high wave from a video sequence of a coastal camera. This detection is performed directly by measuring wave contour in each image using Active Contour Models and by tracking the contour. Since our method measures and tracks the wave contour, it has the following advantages: 1) it is robust to other moving objects; 2) it can be applied to occurrence prediction of overtopping wave. In experiments using real video sequences including both of typhoon approaching and tranquil scenes, we confirmed the effectiveness of our method.
  • 山口 友之, 橋本 周司
    2007 年 127 巻 4 号 p. 605-614
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    Crack width is an important data in the inspection and diagnosis for the safety management of concrete structures. Usually the crack width measurement is done manually by specialists using crack scale. However, it takes a long time and lacks the objectivity for quantitative analysis. Therefore, the importance of image processing for visual inspection has been increasing in civil and construction engineering. Recently, many methods for the crack width measurement have been proposed. We have already tried to measure crack width utilizing image processing. However, these methods concentrated only on the application to test specimens. In this paper, we proposes an improved image measurement method of crack width for the application to real concrete structures. Our method attached crack scale on the concrete surface in the image acquisition. Crack width is measured by the brightness of crack scale areas on the image with sub-pixel order accuracy. In the real world, the lighting condition is different in various environments. For its difference, this paper clarifies the efficient condition of our method for the practical use. The validity of the proposed technique is investigated through experiments with images of a real concrete surface.
  • 北明 靖雄, 奥田 晴久, 橋本 学, 金子 俊一
    2007 年 127 巻 4 号 p. 615-622
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    This paper describes an improved method based on DAI(Depth Aspect Image) matching and HM-ICP (Hierarchical M-ICP) which has been proposed as robust, fast and highly accurate 3D object recognition of model and scene data. DAI matching is robust and fast for objects in a cluttered scene and HM-ICP is fast and of high accuracy but requires good starting point for models. Our approach to object recognition with 3D model recognition procedures is a two step procedure. In the first step, models are searched for in a scene with DAI matching and in the second step, these models are refined by the Hierarchical M-ICP algorithm. Some experiments for real objects show the effectiveness of the proposed method.
  • 泉 哲也, 服部 哲郎, 舒 清宇, 北島 博之, 山崎 敏範
    2007 年 127 巻 4 号 p. 623-629
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    A method for face recognition of acquaintance in a given people image with general background is proposed. We assume that the face pattern to be recognized in the input image is perturbed approximately by Affine transform (rotated, enlarged or reduced, and translated) from a registered original one and also by a lighting variation. In order to estimate the parameters of the Affine transform, the method uses a Karhunen-Loeve (KL) expansion, spatial correlation, and a correction equation based on Taylor's expansion of the transform. Moreover, in order to be insensitive to the lighting variations, we deal with a normalized gradient vector field (NGVF) pattern representation. The experimental result shows that the method using NGVF representation and approximate parameter estimation is considerably effective.
  • 高木 裕治, 中垣 亮, 宍戸 千絵, 田中 麻紀
    2007 年 127 巻 4 号 p. 630-635
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    The cross-sectional dimensions of the photoresist in the photolithography process are estimated from top-down scanning electron micrographs for improved control of photolithography conditions. As semiconductor design rules shrink, strict process control is becoming increasingly important. Two primary process parameters in the photolithography process, exposure dose and focus position, require strict control in order to achieve the desired photoresist profile. The relationship between the photoresist profile and changes in these two parameters of photolithography is determined by simulation, and features suitable for estimation of the photoresist profile are identified in conventional images obtained for critical dimension monitoring. Experimental results demonstrate that process parameters can be estimated to within error of 1.0 mJ/cm2 for exposure dose and 80 nm for focus position, making the proposed method suitable for photolithography process control.
  • 渡邉 恵理子, 石川 さゆり, 太田 真衣子, 小舘 香椎子
    2007 年 127 巻 4 号 p. 636-643
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    We propose a high security facial recognition system using a cellular phone on the mobile network. This system is composed of a face recognition engine based on optical phase correlation which uses phase information with emphasis on a Fourier domain, a control sever and the cellular phone with a compact camera for taking pictures, as a portable terminal. Compared with various correlation methods, our face recognition engine revealed the most accurate EER of less than 1%. By using the JAVA interface on this system, we implemented the stable system taking pictures, providing functions to prevent spoofing while transferring images. This recognition system was tested on 300 women students and the results proved this system effective.
<制御・計測>
  • 渡辺 隆, 草野 洸, 藤原 孝幸, 輿水 大和
    2007 年 127 巻 4 号 p. 644-651
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    3D image processing algorithms for real time calibration of the base line length in single camera stereo system are presented in this paper, and the performances of less than 0.03mm depth measurement in precision and about 0.45sec processing time in the inspection are also presented based on the applications to the experimental equipment. It is technically noted that the base line of this single camera stereo was precisely calibrated by the image processing procedure.
    The scope of application of this simply configured stereo system could be expanded to the wider fields of the industrial 3D measurements.
  • 渡辺 隆, 草野 洸, 藤原 孝幸, 輿水 大和
    2007 年 127 巻 4 号 p. 652-658
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    In this paper, the robust 3D image measurement method is proposed to suppress the occlusion by proposing a quasi Multi-Baseline-Stereo method based on a single camera configuration. We have no way to suppress completely the “occlusion problem" in the field of stereo geometry. In order to cope with this problem, multi-baseline-stereo configuration is often adopted and basically effective. This paper proposes a new configuration of the stereo measurement my means of a single camera, mechanical feeder of the object, image processing for base line calibration, and especially a diagonal feeding of the object in front of the camera. The concept of this method is to realize a pair of stereo geometries on the stereo configuration of a single camera and a mechanical feeder. It is technically noted that the base line of this single camera stereo was precisely calibrated by the image processing procedure in both horizontal and vertical directions.
    This proposed system with the simple hardware composition could be applicable to several kinds of 3D measurements in several industrial fields.
<システム>
<ソフトウェア・情報処理>
  • — 印象評価のポジショニング分析 —
    藤澤 隆史, 土屋 晋, 高島 杏菜, 原田 甫, 長田 典子
    2007 年 127 巻 4 号 p. 667-673
    発行日: 2007/04/01
    公開日: 2007/07/01
    ジャーナル フリー
    In this study, we evaluated observers' impressions on different styles of make-up applications. We showed the observers various styles of make-up for 3DCG characters, and they scored their impression of the made-up faces. In experiment 1, to identify the impression dimensions for the made-up faces, we had 38 undergraduate subjects evaluate 4 make-up styles using 23 adjectives. As a result, we extracted 2 factors, "external appearance" and "internal feeling" about the characters. In experiment 2, we had 70 undergraduate subjects evaluate 5 make-up styles using 7 adjectives. We applied the “positioning analysis" method for the SD data, and mapped the loading values for make-up styles (targets) and adjectives (scales) on a 2 dimensional plane. The results showed that the adjectives were different for each make-up styles, suggesting that the impression of CG characters can be controlled by make-up using texture synthesis.
feedback
Top