IEEJ Transactions on Electronics, Information and Systems
Online ISSN : 1348-8155
Print ISSN : 0385-4221
ISSN-L : 0385-4221
Volume 115 , Issue 3
Showing 1-25 articles out of 25 articles from the selected issue
  • Noriaki Muranaka, Shigeru Imanishi
    1995 Volume 115 Issue 3 Pages 337-344
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Because the gray and colored images have multiple levels, a multiple-valued logic system for picture processing is more effective as compared with a binary logic system.
    This paper proposes several operations and presents ternary digital picture processing circuits (TDPPC's). The TDPPC's consist of ternay/binary level converters, binary/ternary level converters, ternary comparators, neighborhood operation circuits, ternary half adders and center pixel output circuits.
    Algorithms for noise reduction, border line extraction, logic difference, thinning, and connected component extraction are used in the developments and designs of the TDPPC's.
    Moreover, this paper describes a basic system including a host personal computer. TDPPC's utilized in this system are verified by the algorithms.
    Download PDF (1956K)
  • Toshihiro Furukawa, Tsunenori Maeda, Hajime Kubota
    1995 Volume 115 Issue 3 Pages 345-353
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In adaptive signal processing, adaptive algorithms hold the important positions. There are mainly both the rapid convergence characteristics and the reduction of computational requirements as the essential items which are demanded to adaptive algorithms. The algorithms, based on the orthogonal projection onto the subspace spanned with the plural input signal vectors, are known as a method to satisfy the requirements above. The orthogonal projection algorithms result in solving the linear equations and the solutions of equations can be represented with Moore-Penrose type generalized inverse matrix. It is important to realize efficiently the orthgonal projection algorithms which include the inverse matrix above. For this problem, the algorithm has been already proposed which applies the Conjugate Gradient Method, called CGM-BOPA. The convergence characteristics of the CGM-BOPA, however, may degrade down when the recursive procedure of the CGM-BOPA are stopped midway due to some affairs, for example, the limitations of the hardware construction etc. Well, this paper presents a new recursive adaptive algorithm which can efficiently perform the orthogonal projection algorithms. Since the proposed algorithm is based on the orthogonal projection onto the direction vectors for the adjustment of the filter's coefficients at any step in one data block, the convergence characteristics of the proposed algorithm are prior to those of the CGM-BOPA if the recursive procedures are stopped midway.
    Download PDF (1294K)
  • Shigeaki Fukuda, Toshihisa Kosaka, Sigeru Omatsu
    1995 Volume 115 Issue 3 Pages 354-360
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Bill money classification has become important due to progress of office automation. In this paper, we consider the bill money classification problem using time series data which have been obtained from Japanese bill money. Especially, we treat three kinds of Japanese bill money such as ¥1, 000, ¥5, 000, and ¥10, 000. For modelling the time series data, we adopt two methods. One is to use Auto-Regressive (AR) models and the other is to use neural networks. First, we make an AR model according to each bill money stated above. Using AR models for three kinds of Japanese bill money, untrained time series data of bill money is classified into one of three categories such that the prediction error is minimized. Then neural networks for prediction of the time series is used to classify the bill money. Furthermore, in order to compare these methods, we introduce a reliability measure for classification and discuss the ability of pattern classification.
    Download PDF (1360K)
  • Masakazu Yokota, Toshihisa Kosaka, Sigeru Omatsu
    1995 Volume 115 Issue 3 Pages 361-366
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In this paper, we propose a bill money recognition method of US dollar using a neural network. US dollar is classified into one of seven categories like 1, 2, 5, 10, 20, 50, and $100 by processing human picture contained in the US dollar. Furthemore, we transform the US dollar data into frequency domain by FFT and use amplitudes of Fourier coefficients as input data of the neural network. Recognition rate and error probability for a checking data set consisting of US dollars different from the training data set are discussed for the proposed methods.
    Download PDF (1711K)
  • Naoyuki Kanou, Michio Inoue, Yasuhiro Kobayashi, Takashi Kawakami, Kim ...
    1995 Volume 115 Issue 3 Pages 367-372
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    We have developed several Communication Aids (CAs) for serious patients (e. q., Spinal Cord Injured Patients, Amyotrophic Lateral Sclerosis etc.) who lose their ability of communication. The input sensor of the CA is very important to use it smoothly. Most patients have used sensors of put-on-type. But it is difficult to put such sensors on their relevant muscle exactly.
    In this paper, we apply the winking to an input signal of the CA. The winking, whitch is detected on a TV camera, is recognized by training a neural network (NN) with three layers. To implove their ability of pattern generalization, we set the initial weights of the NN refering to its input images.
    Download PDF (1480K)
  • Tomohiko Sato, Hirohide Ushida, Toru Yamaguchi
    1995 Volume 115 Issue 3 Pages 373-380
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    In order to realize creative thinking supports for image generation, this paper proposes an image generation method driven by chaotic dynamics using facial expressions model constructed on fuzzy associative memories. This proposed method supports user's creative thinking for facial expressions and the method clarifies his image by repeating the following procedures.
    (1) An user is shown a candidate for facial expression by chaotic retrieval on fuzzy associative memories.
    (2) The user selects a candidate which pleases him, and change the external input to the network.
    Computer simulation results for actual facial expression data show that the method is effective for creative thinking supports. Further, a leaning method is shown, by which facial expressions obtained by creative thinking supports are memorized.
    Download PDF (1313K)
  • M. Nishida, T. Otsuka
    1995 Volume 115 Issue 3 Pages 381-388
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper presents the results of using the fuzzy regression model to analyze the water quality from satellite images. A case study has been conducted and the results of the fuzzy regression model have been compared with the correlation coefficient of the conventional statistical model. To draw the possibility distribution map of the water quality, a new method with the fuzzy level-slice is proposed in this paper.
    It is shown by statistical analysis that the CCT count number of the remote sensing data has no relation to the water quality. However, the fuzzy regression analysis shows that the CCT count number is related to two parameters about the water quality. They are SS (Suspended Solid) and T-P (Total Phosphorus). Therefore, it is indicated that the fuzzy regression model is applicable to the analysis of the water quality.
    The estimation map of the water quality for SS has been drawn by using the fuzzy level-slice method. It is shown that the pollution level of the estimation map corresponds with the forecasted range from the actual data of the water quality. It also becomes clear that the distribution of the pollution level on the estimation map corresponds with the knowledge of an expert in the investigation of the water quality. Therefore, it can be concluded that the fuzzy level-slice method is a useful technique for drawing the estimation map of the water quality.
    Download PDF (2263K)
  • Kazuo NAKAZAWA
    1995 Volume 115 Issue 3 Pages 389-395
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper describes a three dimensional vision sensor which fuses a sparse range data obtained by the FG vision sensor and 2-D intensity images by a CCD camera of the sensor.
    The range image obtained by the sensor is precise, but coarsely sampled data. Therefore, we cannot recognize a 3-D object densely. On the other hand, we can get features (such as outlines, edge lines, vertices, and so on) from 2-D intensity image, because intensity images are high density data equivalent to the camera resolution. However, we cannot determine their location on 3-D space without some restrictions from only intensity images. To solve this problem, we propose the method that can recognize 3-D Polyheral object precisely and densely by fusing both information obtained from sparse range image and abstracted 2-D image. As a result, this vision sensor is small and light enough to be fitted to a robot hand.
    Download PDF (2945K)
  • Keiichi Yamada, Tomoaki Nakano, Shin Yamamoto
    1995 Volume 115 Issue 3 Pages 396-402
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Research and development on autonomous vehicles which move indoors with a vision system has been actively carried out. At present, however, the autonomous vehicles in practical use are capable only under good lighting conditions. One of the reasons for the limited condition is the narrow dynamic range of TV cameras which input images into the vision system.
    We have developed a method for expanding the dynamic range of TV cameras. A trial vision sensor with a wide dynamic range has been manufactured using the method.
    The developed method makes it possible to obtain an image with a wider dynamic range than that of TV camera itself by combining the images taken under different exposure conditions. A dynamic range of 104 was obtained by the developed sensor. The resolution, frame rate and sensitivity of the developed sensor were almost the same as those of conventional TV cameras. The developed sensor was applied to the vision system of an autonomous vehicle; The effectiveness of the method has been confirmed by the experiments under various lighting conditions.
    Download PDF (3872K)
  • Atsushi Nomura, Hidetoshi Miike, Etsuro Yokoyama
    1995 Volume 115 Issue 3 Pages 403-409
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    The gradient-based method is known as a representative method determing optical flow caused by relative motion between a TV camera and scene from sequential images. The method is based on a basic constraint equation connecting partial derivatives on spatio-temporal brightness distribution to a motion vector. In this paper, a generalized basic constraint equation is derived from observation of temporal change of total brightness in a fixed small region. The equation includes terms of a motion vector, diffusion of brightness and rate of brightness generation. The diffuse term describes physical phenomena of diffusing colored liquid or chemical reactions with diffusion. In addition blurring phenomena in changing focal length of a camera can be represented by the diffuse term. Consequently, we expect that consideration of the diffuse term solves the problem of the non-rigidity of a brightness pattern which is caused by the diffusion. Thus the method assuming the additional following conditions; constancy of diffuse coefficient and of motion field in a small region and no brightness generation is realized. Through analysis of artifical and real blurring image sequences the usefulness of the method is confirmed.
    Download PDF (4582K)
  • Keiichi Uchimura, Takanobu Hanaoka, Takashi Yatuda, Yoshitaka Kiyama, ...
    1995 Volume 115 Issue 3 Pages 410-416
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper describes an image segmentation method using both chromaticity and brightness information to obtain general components of image. The chromaticity distribution of the image is represented by a tree, thereby chromaticity space is split. The merging regions of the image are created by means of simplification of the tree representation and then each region is segmented by brightness information. Classifying the color image as between high brightness pixel with hue or the other, the proposed method can decrease an information of noise included under low brightness pixel. The process time of the proposed method is very shorter than that of the existing method. Furthermore, a mutual relationship between segmentation in each region that can not be obtained using only brightness information can be obtained using both chromaticity and brightness information.
    Download PDF (2242K)
  • Dingding Chang, Shuji Hashimoto
    1995 Volume 115 Issue 3 Pages 417-422
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    The Hough Transform is an elegant way of extracting global features like parametric curves from binary edge images, however, its large computation and memory requirements prevent it from being used for practical computer vision tasks. This paper presents a new approach to detect parametric curves using the Inverse Hough Transform. The key idea of this method is to make the voting process on the image space instead of that on the parameter space in the conventional method, then convert the local peak detection problem into a parameter optimization problem. This leads to substantial saving, not only in storage requirements but also in the amount of calculation required. The experimental results and qualitative analysis showed that in comparison with the conventional Hough Transform methods, the new method has advantages of high speed, small storage, arbitrary parameter range and high parameter resolution.
    Download PDF (3183K)
  • Kenji Shoji
    1995 Volume 115 Issue 3 Pages 423-429
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    As a function of vision system, it is important to identify each object and decide its pose in a scene. When the scene is the silhouette of planar objects overlapping each other, and when the shapes of objects to be identified are given as models, the sequence of these two tasks is called 2-dimensional partial shape recognition. In this paper, we propose a novel method for partial shape recognition. The proposed method uses information of curvature and tangential line direction at each points on digital contours of model objects and a scene. The pose decision algorithm is based on the concept of generalized Hough transformation. To reduce the execution time, contour points data lists are sorted by their curvature. To increase the reliability of results, contour points of models are re-sampled densely in high curvature segments and coarsely in straight line segments. Moreover, to reduce the memory space for generalized Hough transformation, coarse-to-fine analysis is applied. Experimental results show that the proposed algorithm provides valid results of partial shape recognition and takes less than 20 seconds per one model object for recognition on a typical workstation.
    Download PDF (1067K)
  • Toshihiko Tsukada, Osamu Ozeki, Shin Yamamoto
    1995 Volume 115 Issue 3 Pages 430-437
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    A method for inspecting solder dip coating condition of a DIP IC has been developed. In this method, solder dip coating conditions are detected from the reflection intensity of the lighted surface of a lead. For accurate detection, a vision system has been constructed which regards the difference in the inclination of the solder surface as the difference in the intensity of the reflected light.
    The system consists of a dichroic mirror and two linear sensors having sources of light which different wavelength for rapid image detection. An IC lead is simultaneously illuminated from the upper left and upper right directions, and the linear sensors take the image of reflected light in each direction. The areas and distributions of the incline and flat regions of the lead are calculated from the intensity of the reflected light.
    Using the developed vision system and new inspection algorithm, a prototype inspection system has been developed. The experiments using the prototype system were performed with 1600 samples, and the efficiency of this method has been demonstrated.
    Download PDF (3300K)
  • Masaru Amano, Masayoshi Kakikura, Osamu Okamoto
    1995 Volume 115 Issue 3 Pages 438-444
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper desciribes an evaluation method for the removed quantity of wall coating removed by a robot which has a sucking disk, a rotatable water jet nozzle for removing work, and four driving wheels. The states of removed wall surface can be classified into three classes, that is, standard surface (S surface), over removed one (O surface) and under removed one (U surface). The wall surfaces are classified using the data represented in HLS space. It seems that each surface has following features; S surface is the concrete surface that looks like smooth surface. O surface shows some aggregates in concrete, which seems coarse surface. U surface has non removed coating area. Images from a CCD camera fixed on the robot can be classified into these groups based on a newly devised method. The method are using follwing values, (a) the color difference between pixels in HLS space, and, (b) percentage of non removed area in processed image.
    Download PDF (3936K)
  • Masafumi Uchida, Hideto Ide, Syuichi Yokoyama
    1995 Volume 115 Issue 3 Pages 445-451
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    It has been considered that electromyogram (EMG) detected by surface electrodes attached to the subject's arm is useful for controlling a robot arm. Moreover, since the amplitude of EMG is changed by physical strength, it is considered EMG possesses effective informations to assume the grip. However, in order to control the robot arm at will, subject have to train their muscles to consistently generate EMG needed to control the robot arm. EMG pattern recognition system is easy to be influenced by the position slip of electrodes and the artifact. In this study, 1/3 octave-analyzed EMG patterns were classified by neural networks which possess learning ability and deal with Interval-Valued data to cope with the position slip of electrodes. Moreover, the grip were assumed by fussyu inference. Applying their results, the robot arm was controlled to adapt to dynamical arm movement. Interval-Valued data is a method express an attribute as a dot in the multi-dimension. For example, the attribute is not constant and is changing. EMG were measured under folloing conditions; (1) closing hand, (2) openning hand, (3) bending wrist to the bending side, (4) bending wrist to the stretching side, (5) turning wrist to the inside, (6) turning wrist to the outside.
    In the experiment of assumption of the grip, EMG generated when subject gripped ‘hand-grip’ were used. Applying their results, the robot arm was controlled. It took a little under 2 minute to begin to move the robot arm since subject began to move his arm. Bending angle was set up 10 degrees at 1 operation Consequently, in point of the rate of recognition, the neural network which deal with real-valued data. Moreover, in order to control the robot arm by assuming the grip using EMG, applying fuzzy inference was useful considerably.
    Download PDF (1821K)
  • Saburo Okada, Masaaki Imade, Hidekazu Miyauchi, Tetsuhiro Sumimoto, Hi ...
    1995 Volume 115 Issue 3 Pages 452-459
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    A new surface inspection system has been developed to detect various types of flaws and defects and also to discriminate kind and degrees of flaws. To realize such surface inspection, we combined the three dimensional shape measurement through laser beam scanning based on the laser slit light method and the flaws and defects identification through diffraction light pattern analysis by the developed photo detecting sensor. This combination makes it possible to detect locations, sizes, kind and degrees of flaws and defects on undulating or stepped planar metal surfaces. The results of experiments, which were performed to verify the effectiveness of proposed method, are discussed.
    Download PDF (3755K)
  • Toshihiro Nishimura, Souzaburou Yoshihara, Kazuyuki Nagatsuka, Takeshi ...
    1995 Volume 115 Issue 3 Pages 460-465
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    We evaluated the nervous control by using the analysis of spectrum components and component coefficients of variance (CCV) on diabetic autonomic neuropathy. The R-R interval time series is analyzed by the autoregressive model (AR model). Changing factors of them are states of autonomic nervous system. Diabetic autonomic neuropathy concerned with the autonomic nervous system is discriminated by using the R-R interval time series in frequency domain. Furthermore, we studied whether the circadian oscillation of the autonomic nervous system can be evaluated by analysis of heart rate variability derived from a 24-hour ambulatory electrocardiograms (Holter ECGs) on healthy subjects, and diabetic autonomic neuropathy subjects. The characteristics on circadian oscillation of the autonomic nervous system was distinctly in healthy subjects, and the characteristics on circadian oscillation of the autonomic nervous system disappeared in diabetic autonomic neuropathy subjects. Therefore it was considered that the circadian oscillation of the autonomic nervous system can be evaluated by analyzing heart rate variability.
    Download PDF (1514K)
  • Morinobu Endo, Kyoichi Oshida, Takuma Kushihashi, Kiyoharu Kobori, Mic ...
    1995 Volume 115 Issue 3 Pages 466-474
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Microstructure of mesophase pitch-based carbon fibers with high strength and high modulus has been studied by using scanning electron microscope. It has been shown that there is a qualitative correlaion between the texture of the fiber cross-section and the tensile strength of the fibers. And it is difficult to estimate the cross-section morphology quantitatively, although the electron microscopy pictures include many information. In the present paper, image analyzing process has been applies to the high resoultion scanning electron microscopy (SEM) pictures on the newly developed high strength pitch-based carbon fibers. It has been suggested that the cross-section morphology of the fibers has important contribution to the tensile strength of the fibers, and many effort has been paid to characterize the cross-section morphology of the fibers. The work station based image analyzer is applied to characterize and evaluate the cross-section structure of the fiber which has correlation with the strength. The digitization method of the SEM pictures of two mesophase pitch-based carbon fibers and computer analysis of the digitized pictures are demonstrated. Front operations of emphasis of contrast, noise reduction binary operaion and thinning, gave a vivid description of the folded structure of carbon layers so-called fibril in the cross-section of the fibers. The structures of the fibers can be characterized quantitatively with tree methods, the measure of the radius of curvature of the fibril, the 2-dimensional fast Fourier transform (FFT) analysis and fractal analysis of the cross-section structure. The results agree well with the tensile strength of the fibers.
    Download PDF (5785K)
  • Hidenori Itoh, Makoto Ikeda, Yasuhito Takeyama, Kohki Satoh, Yoshitaka ...
    1995 Volume 115 Issue 3 Pages 475-480
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    A computer-aided spectroscopic system by use of an image processing method has been developed in order to diagnose various processing plasma. This sytem is composed of a monochromator, two image intensifiers, a CCD camera and a computer. This is characterised by the employment of a digital signal processor and spatial filtering treatment of image data for fast image operations. The validity and the significance of this system have been confirmed by the results of the spatiotemporally resolved optical emission diagnostics of plasmas in H2.
    The time-resolved spectra have been obtained clearly and the time variation of the profles has been explained from the viewpoint of the self-bias voltage in plasmas.
    Download PDF (1399K)
  • Toshio Asano, Jun Mochizuki, Takashi Ohta, Nobuo Fukuhara
    1995 Volume 115 Issue 3 Pages 481-487
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    This paper presents a method for measuring and evaluating electron beam profiles of high resolution color CRTs in high speed. The system has a two-dimensional sensor to get the intensity data of an electron beam. The magnetic field is applied from the front panel side to shift the electron beam, and used to get the intensity data of the electron beam which is blocked by the shadow-mask. In this system, electron beam profiles (6 points×3 colors) can be measured within 10 minutes for each CRT. The beam profiles are then used to evaluate beam focusing grade. Some parameters (beam area, diameter, vertical and horizontal ratio etc.) are calculated to characterize beam profiles. Human evaluation was executed by a skillful inspector, who judged the beam focusing grade by the readability of character patterns on CRT screens. As a result, we found that two segmented areas' values (3% and 30% of profile intensity maximum) which represent a halo characteristics and a core size, respectively, have strong relations to human evaluation results.
    Download PDF (3331K)
  • Hideo Hasegawa
    1995 Volume 115 Issue 3 Pages 488-495
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    To improve the characteristics of a DC thin film EL device, it is required to make a high crystalline phosphor film. In this paper, by using a successive vaccum deposition method, the fabrication conditions of the ZnS (Cu, Mn) thin film which has a high crystallinity are investigated. The ZnS (Cu, Mn) film is made by successively depositing ZnS (1), Mn, Cu and ZnS (2) on a glass substrate which was coated with SnO2 electrode. The diffusion of Mn, Cu into the host ZnS is carried out by increasing the substrate temperature to 580°C during deposition of the ZnS (2) film. In order to grow the ZnS (2) film on the ZnS (1) film epitaxially, a highly oriented ZnS film as the ZnS (1) film is used. After diffusing most of Mn, Cu into the ZnS (1) film, the ZnS (2) film is deposited. It is found that the ZnS (1) film deposited under keeping the substrate at 300°C has best crystal orientation along <111> direction. Therefore, the ZnS (2) film grown along its direction is made and the highly oriented ZnS (Cu, Mn) film is obtained. With this ZnS (Cu, Mn) film, we have succeded in fabricating the DC thin film EL device which has a luminous efficiency of 0.92 lm/W at the luminance of 1, 300 cd/m2 and which can continuously drive a luminance of 330 cd/m2 for 1, 100 hours (half-life of 3, 400 hours).
    Download PDF (2566K)
  • Toyohiro Kitada, Masafumi Hagiwara
    1995 Volume 115 Issue 3 Pages 496-502
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    When the inputs to fuzzy inference system are incomplete, it is very difficult for the system to infer correctly. For example, even if only one input is lacked, suitable rules cannot be selected. In this paper we propose a new method of pattern matching, named “FIDP (Fuzzy Inference Dynamic Programming) matching” to solve such problems. Since the FIDP matching method utilizes the conventional Dynamic Programming method, it is robust for incomplete inputs. To demonstrate the effectiveness of the proposed FIDP matching method, gesture recognition experiments have been carried out.
    It has shown that in case of the conventional fuzzy inference method (truth space approach) the recognition rate was 75.0%, whereas in case of the proposed FIDP matching method, that was 91.7%.
    Download PDF (925K)
  • Seiichi Serikawa, Teruo Shimomura
    1995 Volume 115 Issue 3 Pages 503-504
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (1150K)
  • Ken Harada, Masafumi Hagiwara
    1995 Volume 115 Issue 3 Pages 505-506
    Published: February 20, 1995
    Released: December 19, 2008
    JOURNALS FREE ACCESS
    Download PDF (157K)
feedback
Top