Reports of the Technical Conference of the Institute of Image Electronics Engineers of Japan
Reports of the 280th Technical Conference of the Institute of Image Electronics Engineers of Japan
Showing 1-25 articles out of 25 articles from the selected issue
  • Mei KODAMA
    Session ID: 16-03-01
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Cached video content management methods with multiple quality (CVCMM-MQ) by using priority estimation have been studied in video content delivery system based on cache type(VCDS-CT). However, they have some practical problems of decreasing the efficiency for the change of the access. Then, the improvement method of CVCMM was proposed for them. In this study, considering decrease of the priority order due to the estimation errors by using estimation methods in the priority values, a new estimation method is proposed to improve the cache efficiency. The adaptive method with the approximation estimations is considered. By the simulation experiments, that the efficiency in the proposed method is evaluated, and it is revealed that the mechanism for decreasing the cache efficiency due to the estimation errors.
    Download PDF (202K)
  • Yoshiharu MASUDA, Hiroshi NAGAHASHI
    Session ID: 16-03-02
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Recently, Mobile phones have been widely spread and used for taking pictures and browsing images. On the other hand, researches about 3D image processing have been developed and lots of 3D contents are widely used now. However, these contents need a special equipment or operations. Therefore, our purpose is taking and browsing 3D images by means of a way close to the current usage of mobile phones with monocular cameras and display. First, we propose a depth estimation method from two consecutive images, where one is taken with flash and another with no flash. We improve the accuracy of depth estimation by noise reduction and intensity difference. In addition, we propose an appropriate display method of 3D images using mobile phones, through subjective evaluation experiments. Then, we make a 3D image system based on our depth estimation and display method.
    Download PDF (24070K)
  • Nobuhiko ONOSE, Kazuma SHINODA, Madoka HASEGAWA
    Session ID: 16-03-03
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In our preceding study, we proposed a graphical password using hybrid images. Visibility of the hybrid image depends on the viewing distance and the structural similarity of the image. In addition, the visibility of the foreground image is different depending on the combination of the foreground image and the background image. Therefore, not all hybrid images are suitable for the graphical password. Takahashi et al. proposed an objective evaluation score using Speeded-Up Robust Features (SURF) for choosing appropriate hybrid images. However, their score does not sufficiently consider the visibility of the foreground image for the legitimate user. In this paper, we proposed an objective evaluation score considering the visibility of the foreground image for both the legitimate user and the shoulder-surfer. Experimental results show that our score can choose hybrid images suitable for user authentication.
    Download PDF (1060K)
  • Ken-ichi TANAKA
    Session ID: 16-03-04
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    This paper, we descrive a part of the study on digital holography. This paper contains the optimization method focusing on improvement of image reconstruction, and the data hiding using digital holography.
    Download PDF (1000K)
  • Daichi Arai, Hiroshi Nagahashi
    Session ID: 16-03-05
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In this research, we propose a navigation system which can estimate a position on a map by aligning a 3D reconstruction result by Visual SLAM on the map. First, the system extracts walls of the building using sight line connecting the cameras and the points. Based on the Manhattan-World hypothesis, building walls are transformed into orthogonal line segments using two orthogonal directions in the building. We generate the floor plan of the building by solving the optimal path problem of the graph with those line segments as edges. By performing the same processing on the map image and comparing the two floor plans, the system aligns the 3D reconstruction result and the map by proposed geometric hashing method.
    Download PDF (5673K)
  • Yoshiki MIZUNO, Takeshi KOBAYASHI, Shinji IKARUGA, Akira OHGAI, Kazuki ...
    Session ID: 16-03-06
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Since walk-through simulator based on virtual space used for evacuation drill provides too much degree of freedom to move in the virtual space, if passable path in a road becomes more complicated its beginner user often encounters a serious trouble in moving through it, he/she can not move the rood as intended. This problem can be solved by setting only one path per every rood section and by enabling the user to move in the virtual space with several simple operations such as forward-moving and stop. On the other hand, considering about progress of aging society, evacuation with a handicapped person sitting in a wheelchair needs to be assumed. In this case, the size of the moving-object should be taken into account. In this paper, a method for an automatic generation of paths on a road with scattered rubbles in which both size of the moving-object and its moving-attitude are considered is proposed.
    Download PDF (648K)
  • Kazuki FUJIHASHI, Kiyotada WASHIYA, Hiroshi NAGAHASHI
    Session ID: 16-03-07
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    This paper proposes a new method for classification of the lung poorly differentiated Adenocarcinoma and nonkeratinizing type Squamous Cell Carcinoma. It’s difficult to distinguish these two types of lung cancers. However, definite histopathological diagnosis is needed for the effective cancer treatment. Firstly, the cell nuclei are detected from the cytological images using a classifier trained by convolutional neural network (CNN). Second, the cell nuclei are distinguished using the classifier trained by SVM and CNN. We compared the experimental results of SVM and CNN. Experimental results show our proposed method can distinguish two types of the lung cancers well.
    Download PDF (673K)
  • Yota ISHIKAWA, Kiyotada WASHIYA, Kota AOKI, Hiroshi NAGAHASHI
    Session ID: 16-03-08
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    The number of cytotechnologist who can diagnose brain tumor is not sufficient, because it is highly specialized. Our purpose is to diagnose microscopy images of brain cortex and identify brain tumor by medical image processing. Computational image analysis may dissolve the shortage of experts and support objective pathological examinations. In this study, we present a novel method to segment cell region robustly using BING objectness estimation and classify brain tumor using deep convolutional neural networks (CNNs) constructed by deep residual learning. Experimental results showed 99.6% accuracy of the classification and F score of the segmentation is 0.922.
    Download PDF (941K)
  • Kenta SAITO, Kousei KIKUCHI, Takanobu YAGI, Jun OHYA
    Session ID: 16-03-09
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
  • Junichi SUGITA, Tokiichiro TAKAHASHI
    Session ID: 16-03-10
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    The red-yellow-blue (RYB) color model is a subtractive model based on pigment color mixing and is widely used in art education. In the RYB color model, red, yellow, and blue are defined as primary colors. Because of the specifics of the human perception system and computer displays, most computer applications use the red-green-blue (RGB) color model; however, this model is not comprehensible for many people because of its use of additive color mixing. In this paper, we formulate a bi-directional conversion between the RGB and RYB color spaces. Easy-to-use of RYB color model in computer can be achieved.
    Download PDF (1150K)
  • Daichi Tanaka, Keiichi Iwamura, Masaki Inamura, Kitahiro Kaneda, Isao ...
    Session ID: 16-03-11
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Recently, in society developed internet, it become easy that ordinary consumers create a new content by editing the content uploaded on the internet by other user. In contrast, the copyright protection system capable of controlling the editing of the content using the digital signature been proposed. This method is the author can set a signature to control the change, delete, addition, and diversion, also the contents itself can be controlled using electric signature. In this presentation, I present utility of the content protection system that can control editing by using contents created by content editing tool ``MikuMikuDance”.
    Download PDF (987K)
  • Tomohiko GOTO, Takafumi SAITO
    Session ID: 16-03-12
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    When students learn mathematics, some of them tend to learn how to solve problems. Then, they don't extend to the understanding of the essence of the operations. So, the purpose of this study is visualization of linear algebra for educational use and development of interface as the teaching material. The visualization policies are as follows. First, 2×2 or 3×3 matrices are considered transformation matrix to express coordinate transformation in the Euclidean space. Then, lattice on the 2D or 3D space is visualized in the case of the transformation. Based on this way of thinking, formulate presentation methods for various operation, and developed the system as a Web application. Additionally, a questionnaire-based evaluation experiment was conducted. As a result, the effective things as a auxiliary teaching material were provided.
    Download PDF (549K)
  • -Survey for Japanese and French Students-
    Mitsuko NISHIGUCHI, Kazuhisa SENZAKI, Kunio OHNO
    Session ID: 16-03-13
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    This paper discuss the future possibilities and issues of the e-Sports’ diffusion, referring the result of survey which was conducted in Japan and France. The e-Sports is a competition that utilizes electronic media and has the both characteristics as a game and sports. It is popular in North America, Europe and Korea. It is said that a professional team was organized, though people’s attention is very low in Japan comparing with other countries such as France. Based upon the result of survey, it seems that the attitude to the game and sports may be different between Japan and France. However, Olympic and Paralympic may be contributing to the diffusion of e-Sports in future if it’s adopted as an official game. Keyword
    Download PDF (717K)
  • Koji KINOSHITA, Masaharu ISSHIKI
    Session ID: 16-03-14
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Natural images contain many edges with various directions, and artifact images contain many edges with a specific direction. Focusing on this feature, the method that calculates a histogram of the edge direction, and its spectrum, have been proposed. However, in circular artifacts, they included edges in all directions, there was a tendency to classify them to natural images. In this paper, we propose a new feature that can classify circular artifacts. There exits the relevance on the edge direction between the target pixel and its neighbors in locally. Hence, we calculate a co-occurrence histogram in the edge direction and adopt its spectrum as a feature. Moreover, we use the non-linear SVM. The experimental result shows that the proposed method is effective as compared with normal histogram of edge direction.
    Download PDF (964K)
  • Fitria YUNINGSIH, Hiroki TAKAHASHI
    Session ID: 16-03-15
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    The implementation of surveillance systems are mostly not yet equipped with automatic feature for object identification. Since the object of surveillance system is mostly to monitor human activity and it captures human appearance from various angles, this paper proposed serial methods to identify personal appearance in the surveillance video without employing face recognition. The process starts from the object region extraction, image segmentation, and features extraction. The only information to optimize is based on the object appearance that is processed to obtain any possible features, for example clothing, hair and skin color. These features are then used for personal identification.
    Download PDF (644K)
  • Siyu CHEN, Makoto. J . HIRAYAMA
    Session ID: 16-03-16
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    When visually impaired persons make up themselves, low vision people can not see if the mirror is not very close, or some visually impaired persons are difficult to discriminate colors. Thus, a system that supports the makeup of visually impaired persons is needed. In this research, in order to develop a makeup support system for the visually impaired persons, a facial region recognition is made and the recognition rates of the face region are evaluated. The recognition rates of face region, eye regions, lips region were 90% or better. Lip countour extraction was also done.
    Download PDF (853K)
  • [in Japanese]
    Session ID: 16-03-17
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Download PDF (5356K)
  • Naoto Maeda, Zhao Wang, Jun OHYA
    Session ID: 16-03-18
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In recent years, the demand for the system that evaluates human motion is increasing. Studies of recognizing motion are actively being conducted, but not many works on quantitative assessment of human motion can be seen. This paper proposes an automatic algorithm that assesses human action for multiple scoring items by learning evaluation scores for each item. More specifically, to obtain the features, the proposed algorithm performs DCT (Discrete Cosine Transform) to the sequences of the 3D coordinates of human joints. The obtained features are learnt by Support Vector Regression. As a result of the leave-one-out method based experiments that give 10 points as the full score to each action, a conventional method and the proposed method achieves the mean error of 0.59 and 0.26, respectively. It turns out that the proposed method can improve the evaluation accuracy.
    Download PDF (432K)
  • Takuto NAGAOKA, Hiroshi UNNO, Masahiro SUZUKI, Kazutake UEHIRA
    Session ID: 16-03-19
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We evaluated a technique of obtaining the visually perceived position of a virtual object presented in front of the screen of a 3-D display. Interactions between the virtual objects and users’ bodies actually seen by the users require executing them at the visually perceived positions of the virtual objects. Although conventional techniques execute interactions at the positions calculated from binocular disparity, the visually perceived positions of the virtual objects are often different from them. The proposed technique meets the requirement described above by obtaining the visually perceived positions from users’ body movements. In the evaluation, we conducted the experiment in which the accuracy and precision of the visually perceived positions obtained from hitting movements in real time. The results obtained from the experiment indicated that the visually perceived positions obtained with the proposed technique were temporally accurate and precise. We demonstrated that the proposed technique was feasible.
    Download PDF (555K)
  • Hiroki HONDA, Zhao WANG, Jun OHYA
    Session ID: 16-03-20
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Towards the actualization of computer teachers that can supervise guitar plays, this paper proposes an algorithm to track the position of the fretboard from the video of guitar plays. In this paper, we use the Shi-Tomasi method to detect the corners of the fretboard as feature points and track the detected points using the Lucus-Kanade (LK) method. However, during the guitar plays, the performer's fingers frequently overlap the fretboard; thus, the feature points cannot be tracked accurately. Therefore, using the RANSAC algorithm, perspective transformation matrix is obtained from the feature points between the two consecutive frames, so that the fretboard can be tracked accurately. By using the different numbers of feature points 20, 30, 40 for the experiments, the best tracking results were obtained when 30 points were used. At that time, the processing time per frame was 7.3 seconds, and the error of the estimated position of the fretboard was 4.9 mm.
    Download PDF (844K)
  • Chanjin SEO, Satoshi SHIMA, Jun OHYA
    Session ID: 16-03-21
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In modern society, the spread of devices such as personal computers, smart phones and doublet terminals has increased asthenopia such as dry eye. In order to solve this problem, conventional research measured astigmatism from eye-blinks so as to capture sleepiness and fatigue. However, the phenomenon of “eye narrowing” caused by drowsiness and fatigue cannot be explained only by eye-blinks. That is why a new “sleepiness” and “fatigue” detection method that can link the eye-blink to asthenopia is needed. This paper proposes a control model that can represent eye-blinks and eye opening and closing based on responsiveness of the eyelid. Experiments demonstrate that the proposed model can describe the eye situation in continuous facial video sequences accurately.
    Download PDF (1013K)
  • Toshiki FUJINO, Kohei KURIHARA, Yoshitaka TOYODA, Daisuke SUZUKI
    Session ID: 16-03-22
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Currently, a surveillance camera which I can take a motion in the low illumination environment is required. There is the noise reduction algorithm using multi images. It is possible to reduce noise in the low illumination environment without object blurring. But, the conventional technique has a problem misrecognizing a noise as a moving object under the environment with much quantity of noise. Therefore we propose the noise reduction algorithm using noise robust motion detection. This algorithm performs motion detection from input image and delayed image and change the noise reduction amount according to the motion detection amount. As a result, proposed method reduces noise about half compared with conventional method.
    Download PDF (395K)
  • Syogo FUJIMARU, Kazuki AKIYAMA, Toshiaki YAMANOUCHI, Kazuhisa YANAKA
    Session ID: 16-03-23
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Video see-through type head mount displays for mixed reality systems are generally large and expensive. Moreover, they aren’t suitable for outdoor use because they should be connected to personal computers. Therefore, we developed a new smartphone-based MR system, in which a 3D prism for taking a stereo video was attached to the camera of the smartphone, and the obtained video was projected on a plane in Unity’s scene. Two character models were placed in front of the plane to produce parallax. The scene was taken with a virtual camera and displayed on the LCD of the smartphone. When the smartphone was housed in VR goggles, the real world and the virtual world both of which had binocular parallax, looked mixed.
    Download PDF (1099K)
  • Yushi MINEMURA, Hideki TOUDOU, Masanori KAKIMOTO
    Session ID: 16-03-24
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In digital animation and video games, peeling marks of adhesive seals are often used to depict our daily life. In this paper we propose a visual simulation method for adhesive seals that can reproduce the irregular geometry of peeling marks. The seal layer structure is modeled by a cubic lattice as elastic thin layers. Lattice vertices are connected by a mass-spring system with a modified Hooke’s Law, which is extended to peeling behavior of adhesive seal. A set of spring constant and breaking point is adjusted for each spring, depending on the spring directions and the adhesive layer thickness. As the result of the proposed simulation in various parameter settings, our method can reproduce plausible geometries for peeling marks.
    Download PDF (846K)
  • Masahiro Suzuki, Keigo Takazawa, Kousuke Nakamura, Youichi Takashima, ...
    Session ID: 16-03-25
    Published: 2017
    Released: June 01, 2020
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We evaluated our previously proposed technique of invisibly embedding information inside a 3-D printed object by using two color 3-D printing and of nondestructively reading the embedding information from the object. In this technique, information is first embedded in the digital data for 3-D printing. Then, the information is also embedded inside a 3-D printed object when the object is fabricated with the data and a 3-D printer. The information inside the object is expressed by constructing the inside with fine domains different in physical characteristics from the main body. The structure of the fine domains inside the object is detected by utilizing nondestructive inspection, and the information is decoded from the detected structure. In the experiment, the main body was fabricated from blue resin that had high transmittance of the near-infrared rays. The fine domains were fabricated from white resin that had low transmittance of the near-infrared rays. Near-infrared transmittance imaging was used to detect the structure of the fine domains inside the object. The results obtained from the experiment indicated that the structure of the fine domains inside the object could be detected by eyesight. We demonstrated the feasibility of the technique.
    Download PDF (282K)
feedback
Top