ITE Technical Report
Online ISSN : 2424-1970
Print ISSN : 1342-6893
ISSN-L : 1342-6893
41.05 Multi-media Storage(MMS)/Consumer Electronics(CE)/Human Information(HI)/Media Engineering(ME)/Artistic Image Technology(AIT)
Showing 1-47 articles out of 47 articles from the selected issue
  • Pages Cover1-
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Download PDF (115K)
  • Pages Cover2-
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Download PDF (646K)
  • Kohei TATENO, Sho TAKAHASHI, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-1
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In order to properly maintain and manage transmission line towers, engineers climbs towers and visually diagnoses paint flm degradation of the towers. Recently, primary inspection using videos captured by a camera which is set at a safe place such as the ground and building roofs has been performed. However, the burden on the inspection by visual inspection is enormous. Therefore, new technologies for transmission line tower inspection using the video are necessary. In this paper, we propose tower region detection method as an initial study of inspection support technology using videos. The proposed method can extract tower regions from the video captured by the camera set at the overhead ground wire. The videos include non-subject regions such as sky and trees in addition to steel tower. Furthermore, since the steel tower area is the foreground and the camera is shaking, the tower area moves more than the background area. Therefore, the proposed method remove background area based on color and extract the large motion area. Then the proposed method extracts the steel tower area by integrating the results.
    Download PDF (632K)
  • Shohei KINOSHITA, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-2
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    This paper presents a quantifcation method of relationship between users and musical pieces. Our proposed relationship is obtained by constructing the heterogeneous graph where nodes are users and musical pieces and applying link prediction to the graph. Our previous method recommends musicby monitoring passage distance from users to musical pieces on the graph. However,our previous method focuses only on passage distance on the graphbut does not consider link structure around the target node pairs. Therefore, the proposed method quantifes the relationship between users and musical pieces by using link prediction methods, which are some parts of the graph structure analysis methods. Link prediction methods predict the presence or absence of link between the target node pairs based on link structure. Finally, we can quantify the relationship between users and musical pieces considering link relationship among nodes to realize the successful music recommendation by introducing the link prediction methods.
    Download PDF (323K)
  • Azuma FUJIMOTO, Toru OGAWA, Kazuyoshi YAMAMOTO, Yusuke MATSUI, Toshihi ...
    Session ID: MMS2017-3
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    We have created Manga109, a dataset of a variety of 109 Japanese comic books publicly available for use for academic purposes. This dataset provides numerous comic images but lacks the annotations of elements in the comics that are necessary for use by machine learning algorithms or evaluation of methods. In this paper, we present the metadata of Manga109 including the annotations of “frame”, “text”, and “character”. And then, we analyze the metadata to reveal the tendency of drawing of every author or genre in Japanese comics.
    Download PDF (840K)
  • Naoki SAITO, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-4
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In this paper, we propose a property estimation method using electron microscope images and mix proportions of rubber materials and discuss its performance. The proposed method estimates property of rubber materials based on Partial Least Squares regression using visual and mix proportion features obtained from the electron microscope images and the mix proportions of the rubber material. Then the proposed method enables the accurate estimation considering the internal structures and mix proportions. The experimental results show the performance of the proposed method and shows its discussion.
    Download PDF (338K)
  • Ren TOGO, Sho TAKAHASHI, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-5
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    This paper presents an automatic selection method of representative images for deterioration diagnosis of steel towers. Although representative imageshave been usedin deterioration diagnosisfor reducingvariationsin diagnostic results, these images have been selected manually by experienced inspectors. Thus, it is desired that representative images will be selected automatically and updated adaptively.We propose the automatic representative image selection method employinga Machine learning in this paper. Furthermore, we examine the effect of representative images selected by our method through the experiment.
    Download PDF (617K)
  • Keisuke MAEDA, Sho TAKAHASHI, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-6
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Automatic distress classification of distresses occurring in road structures is necessary in order to support inspectors for maintenance inspection. This paper presents distress classification method using deep learning for improving classification performance. Specifically, the proposed method generates a classifier based on Deep Extreme Learning Machine which is one of deep learning methods, constructs Auto Encoder for each hidden layer and sequentially determines parameters between hidden layers. Consequently, realization of more accurate distress classification is expected compared to previously machine learning methods.
    Download PDF (636K)
  • Yoshihiko KAWAI, Takahiro MOCHIHZUKI, Masanori SANO
    Session ID: MMS2017-7
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    An effective technique of retrieving desired video scenes is necessary to manage huge video archive. Especially at the broadcasting station, there are demands of producers to retrieve video scenes with specific person. Firstly, we need to detect face position at a video frame before recognizing person. But existing face detection methods have an issue that the accuracy decreases when target video includes large variation of appearance such as illumination condition, facial direction and facial expression. This paper proposes a novel face detection method for TV video which is robust to such video variation. The proposed method uses cascaded decision trees and Gabor convolution filter to improve both detection accuracy and processing cost. In experiment, we used broadcast TV video to verify an effectiveness of the proposed method. We also performed comparison experiment with several existing methods to verify superiority of our method.
    Download PDF (254K)
  • Yuma TANAKA, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-8
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In this paper, we present an auditory target detection method using auditory and functional Magnetic Resonance Imaging (fMRI) features calculated from auditory signals and Blood Oxygenation Level Dependent (BOLD) signals scanned by fMRI, respectively. Our method classifies target segments and non-target segments via collaborative use of the two kinds of features. Specifically, our method constructs two classifiers respectively using only auditory or fMRI features and integrates these results based on the confidence output from each classifier. Consequently, multimodal classification becomes feasible by our method. Experimental results show the improvement of our method over the methods using only auditory or fMRI features.
    Download PDF (356K)
  • Daichi TAKEHARA, Ryosuke HARAKAWA, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-9
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    We have previously proposed a Web video retrieval method using hierarchical structure of Web video groups, which are sets of Web videos with similar topics. Although conventional works verified the e ectiveness of the retrieval method on the basis of an evaluation metric for clustering, verification for the usability such as degree of users’ satisfaction was insufficient. Therefore, this paper newly proposes a visualization interface of the retrieval method, and enables a subject experiment that assumes real-world deployment. By the experiment, the usability of the retrieval method can be confirmed.
    Download PDF (746K)
  • Shota HAMANO, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-10
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    This paper presents a method for accurate extraction of concept relationships using tagged images. Previous methods extract concept relationships using either or both of visual features and textual features extracted from tagged images. In the method that we have previously proposed, visual similarity and textual similarity are calculated based on kernel density estimation and word2vec, respectively. Although kernel density estimation considers distributions of the visual features, there is still room for accuracy improvement of concept relationship extraction. In this paper, we utilize locality-constraint linear coding (LLC) to achieve accurate extraction of concept relationships, which is robust to visual variations. The proposed method also utilizes GloVe, which reportedly represents concepts more effectively than word2vec in the field of natural language processing. Experimental results show that LLC and GloVe contribute to effective representation of concepts and improve the accuracy of the subsequent extraction of the concept relationships
    Download PDF (219K)
  • Kurumi KAMINISHI, Terumasa AOKI
    Session ID: MMS2017-11
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Pictogram, which is a simple picture-based symbol, is widely used to indicate important facilities (such as “rest room”) or important rules (such as “no smoking”). A lot of new application will be generated if a practicable pictogram matching method is provided. But it is difficult to use those techniques in practice since pictogram matching by existing shape descriptors takes a lot of time. In this paper, we present novel speed-up techniques of pictogram matching using local shape descriptor. Our method selects the candidates of matching pairs from contour and inner structure information. The experiments show our method improves computation time for existing methods.
    Download PDF (1516K)
  • Yuma SASAKA, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-12
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    This paper presents a method which estimates interest level by quantifying facial expression data captured from users while watching videos. In the proposed method, the framework for anomaly detection is newly applied to estimate interest level. Specifically ,by using a infrared depth sensor, our method obtains changes in facial expression as numerical features, which are obtained from users while they are watching the videos. Next, the probability distribution of the features is calculated to model changes in facial expression. Then, an anomaly score of changes in facial expression is obtained based on the probability distribution. Using the calculated anomaly score, interest level, which the change of facial expression represents, could be estimated. Finally, this paper shows the effectiveness of the proposed method through experimentation on real participants.
    Download PDF (592K)
  • Kenta ISHIHARA, Sho TAKAHASHI, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-13
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In this paper, we investigate deformation detection in subway tunnel using Convolutional Neural Network (CNN). In Japan, appropriate management of infrastructures is currently an important task since the infrastructures constructed intensively on the high economic growth period have aged. In order to realize support technology of the maintenance inspection, we construct the deformation detection method using CNN which enables accurate image recognition. Furthermore, we compare the detection performance between the method using CNN and the method which detects the deformation using visual features.
    Download PDF (2053K)
  • Kento SUGATA, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-14
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    This paper presents a method that estimates human emotion evoked by visual stimuli using functional near-infrared spectroscopy(fNIRS) signals. The proposed method enables estimation of individual emotion based on feature extraction and machine learning methodology. In our method, Fisher score-based supervised feature selection and successively orthogonal discriminant analysis (SODA)-based supervised dimensionality reduction are applied to fNIRS features extracted from fNIRS signals. Fisher score enables selection of effective features, i.e., channel, for estimating human emotion. Then SODA obtains transformed features that consider the relationship between the effective feature and the emotion evoked by visual stimuli. The performance improvement of emotion estimation can be expected by using the obtained features. Experimental results obtained by applying our method to actual fNIRS signals show its effectiveness.
    Download PDF (1256K)
  • Takamasa FUJII, Soh YOSHIDA, Mitsuji MUNEYASU
    Session ID: MMS2017-15
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Video reranking is an effective way for improving the retrieval performance of keyword-based video search engines. A fundamental issue underlying the success of existing video reranking approaches is the ability in identifying potentially useful recurrent patterns from the initial search results. These patterns can be leveraged to upgrade the ranks of visually similar videos, which are also likely to be relevant. However, mining useful patterns without understanding query may lead to incorrect judgment in reranking. We explore the user selected data, which can be viewed as the footprints of user searching behavior, as an effective means of understanding query, for providing the basis on identifying the recurrent patterns that are potentially helpful for reranking. In this paper, a new reranking algorithm, named user feedback assisted multi-modality reranking, is proposed. The algorithm leverages selected videos to locate similar videos that are not selected, and reranks them in a multi-modality learning scheme. Experimental results obtained by applying the proposed method to a real-world video collection show its effectiveness.
    Download PDF (1076K)
  • Mizuki SHIMAKAGE, Azusa FUJITA, Nobuo EZAKI, Mitsuhide ISHIKAWA, Tomoa ...
    Session ID: MMS2017-16
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    According to the domestic estimate of population of the physical disabilities, there are 61,000 visually impaired people who are not in any living support center and able to be hired (aged 18 to 59) in Japan. Recently, the number of employment of the visually impaired in administrative work is increasing. To get such kind of job, it is necessary to gain IT skills. It is desirable that they are trained how to use computer at living support centers for the visually impaired. However, there are only a few support centers and also the centers are mostly around the metropolitan area so there is a disparity among regions. On the other hand, online service is getting not to depend on regions because of cloud services. Therefore, we constructed a study support system with Office365, a cloud service, to help visually impaired people and their supporters who want to study even though they are at home. Also the system is evaluated at an independent living support center for the visually impaired.
    Download PDF (1043K)
  • Azusa FUJITA, Mizuki SHIMAKAGE, Nobuo EZAKI, Mitsuhide ISHIKAWA, Tomoa ...
    Session ID: MMS2017-17
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Gathering information has become easy for visually impaired people due to the screen reader that reads contents on display. However, they need to become skilled at using a keyboard, and knowledgeable about character input. Also, they have to know how to gather information from the Internet. Therefore, a study support web site and two applications to study basic computer skills have been developed. A web site has also been developed to check their progress of study. In addition, an evaluation experiment was done to confirm that learning data is stored on our storage correctly.
    Download PDF (1471K)
  • Yusuke FUKUSHIMA, Toshihiko YAMASAKI, Kiyoharu AIZAWA, Kenshiro MORI, ...
    Session ID: MMS2017-18
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    While massive open online courses (MOOC) have gained increasing popularity in recent years, predict- ing the number of students who attend and leave from classes is an important task to analyze their interests. In this paper, we proposed a method to predict the number of views and dropout rates. The effect of each factor, such as broadcasting dates and contents of classes, was also investigated to reveal the components that are significant to prediction. We tested our method using a collection of 2,327 lecture videos broadcasted in Schoo.
    Download PDF (644K)
  • Ryuichiro GODA, Hiroki TAKAHASHI
    Session ID: MMS2017-19
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    The privacy problem is caused by various factors. First, I will reconsider to organize information and privacy and their handling in Modern. Thereby, the concept of privacy in modem Japanese society, I described that is based on the self-information control rights and relational mobility. Next, I estimated the posture of a person and created a skeleton model. And then I protected the privacy of the subject based on the skeleton model. As a result, the estimation error between the measured skeleton and the estimated skeleton model was 18 cm, and I created an image that converted natural look of the subject.
    Download PDF (799K)
  • Kiyomi SAKAMOTO, Yutaka TANAKA, Kuniko YAMASHITA, Akira OKADA
    Session ID: MMS2017-20
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    We experimentally estimated of psychophysiological state during high and standard dynamic ranges contents viewing on a 58-inch 4K TV. As the measurements items, subjectivity assessment, blinking rates, heart rate variability (the level of sympathetic nerve activity: LF/HF), near-infrared spectroscopic (NIRS) topography were adopted. The results showed that the scores for “presence,” “impact,” “dynamic,” “feeling of depth,” and “high quality,” when viewing 4K-HDR video content were significantly higher than those for 4K-SDR content. Moreover, LF/HF, representing sympathetic nervous system activities, when viewing 4K-HDR video content, was significantly lower than those for 4K-SDR content.
    Download PDF (536K)
  • Makoto TADENUMA
    Session ID: MMS2017-21
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    The author has developed methods to estimate the unpleasantness degree of shaking images which should cause a certain harmful influence on viewers’ body. However, it was found that the estimation error of the unpleasantness degree could not be reduced to a level low enough when the unpleasantness would be directly estimated with the physical characteristics of the shaking images. Therefore in order to precisely estimate the unpleasantness degree, a new psychological index named “cognition degree of shakiness”, which occurs previous to the unpleasantness, was introduced. A series of evaluation experiments were carried, and the differences between the cognition degree of shakiness and the unpleasantness degree were found out to be caused by the influence of “shaking frequency”, “shaking duration”, “shape of shaking area”, and “viewing area size”.
    Download PDF (1523K)
  • Akio KAMEDA, Megumi ISOGAI, Daisuke OCHI, Hideaki KIMATA
    Session ID: MMS2017-22
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Some efficient 360-degree omnidirectional video streaming methods are in the market these days. In general the methods make it possible to stream the video in a limited network bandwidth by sending a part, user’s RoI (Region of Interest), of the video. However, it can impair user’s QoE (Quality of Experience) due to quality discontinuity of the image at the RoI boundary section. In this paper, we propose a method that can mitigate the discontinuity and improve the QoE by smoothing the section. Subjective evaluation results are also shown.
    Download PDF (696K)
  • Kei OGURA, Kodai KIKUCHI, Takeshi KAJIYAMA, Eiichi MIYASHITA
    Session ID: MMS2017-23
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    We developed a compression recorder for full specification 8K Super Hi-Vision and a system which can back up the recorded 8K data from a memory package. To realize fast transferring, we designed the backup system with a memory control board which is used in the compression recorder. We measured data backup speed and found that around 24Gbps data rate between the memory package and the memory control board was achieved.
    Download PDF (485K)
  • Yusuke KANETAKE, Nobuo TAKESHITA
    Session ID: MMS2017-24
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    We propose a stable adjustment method of loop gain for feedback control. In the proposed method, stable adjustment can be realized by being done without feedback control. We applied this proposed method to vibration suppression control of objective lens in optical disc system, and crossover frequency of open-loop characteristics would be a desired one between -30 and +85 degree.
    Download PDF (418K)
  • Yuta GOTO, Atsushi OKAMOTO, Kazuhisa OGAWA, Akihisa TOMITA
    Session ID: MMS2017-25
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In the holographic memory, the achievable multiplexing number of holograms and the recording density are inherently restricted by the dynamic range in a medium. To suppress the consumption of dynamic range in a medium, we propose a multiplexing technique using the virtual phase conjugation. In this method, multiple signals that contain data pages are simultaneously recorded into a recording medium by utilizing the property of the beam diffusion with a random diffuser and the wavefront reconstruction of a phase conjugation light. Consequently, this method enables us to reduce the exposure amount for recording data pages. Furthermore, this method keeps the increasing size of optics at minimum by computationally processing the phase conjugate reconstruction of data pages. In our simulation, by observing the refractive index modulation depth in medium in which holograms recorded, we confirm that this method can suppress the consumption of the medium’s dynamic range. In addition, we also confirm that the suppression for the consumption of dynamic range improves the recording capacity in the holographic memory.
    Download PDF (1084K)
  • Nobuhiro YAMAGISHI, Atsushi OKAMOTO, Jin NOZAWA, Yuta GOTO, Kazuhisa O ...
    Session ID: MMS2017-26
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In phase measurement using Holographic Diversity Interferometry (HDI), misalignment of two holograms captured by two image sensors cause the deterioration of the measuring accuracy. In order to solve this problem, we propose the virtual shifting method for compensating the alignment error of HDI. The proposed method can reduce the measuring error of HDI by virtually shifting the spatial position of either of holograms captured by two image sensors. In this paper, we confirmed the basic operation of the proposed method by an experiment.
    Download PDF (710K)
  • Fumiya MIZUKAWA, Atsushi OKAMOTO, Yuta GOTO, Shimpei SHIMIZU, Kazuhisa ...
    Session ID: MMS2017-27
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Toward expanding the transmission capacity, mode division multiplexing (MDM) transmission using few-mode fibers has attracted attention. For the construction of the network using the MDM in the future, we proposed the spatial mode separation and conversion technique using volume holograms. Our proposal enables us to carry out the spatial mode separation and conversion simultaneously using angularly multiplexed volume holograms with a single device. In this study, we demonstrated the basic operation of our proposal and reported the results.
    Download PDF (1451K)
  • Simpei SHIMIZU, Atsushi OKAMOTO, Fumiya MIZUKAWA, Kazuhisa OGAWA, Akih ...
    Session ID: MMS2017-28
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In the mode division multiplexing transmission, the spatial mode de-multiplexing technique is required on the receive side. The volume holographic mode de-multiplexer (VHDM) provides the mode de-multiplexing function with a single device by utilizing angularly multiplexed volume holograms. However, it is difficult to apply the VHDM to the actual optical communication system because typical holographic mediums have no sensitivity in optical communication wavelength band. In this research, toward applying the VHDM to the optical communication system, we demonstrated the mode de-multiplexing by using the dual-wavelength method which allows us to record the volume hologram with the light of 532 nm in wavelength but use it to separate infrared readout beams of 850 nm in wavelength.
    Download PDF (1211K)
  • Kohei KURIHARA, Yoshitaka TOYODA, Daisuke SUZUKI
    Session ID: MMS2017-29
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    We propose a novel image fusion approach for infrared (IR) and visible (VIS) images. The proposed approach can be applied for many application (e.x. denoise, dehaze) in a single framework. Detail components are extracted from VIS and IR images. Before synthesizing them, these components are enhance/inhibit depend on a scene.
    Download PDF (491K)
  • Ritsuya OSHIMA, Masashige SUWA, Muneharu KUWATA, Kuniko KOJIMA
    Session ID: MMS2017-30
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Recently, a projection lighting that enables more effective information transfer has penetrated and it can project content and information to the wall and the ground. When rotating and translating images, it needs to change the direction/position of images in a area of lighting device such as LCD, DMD and so on, or of the entire lighting equipment. In such a case, it decreases a lighting efficiency or needs a large space to move the lighting equipment. This paper proposes an adaptive lighting system with movable prism mechanism that can decrease a size of lighting equipment.
    Download PDF (640K)
  • Haruhiko OKUMURA
    Session ID: MMS2017-31
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    This report is a summary of the consumer electronics trends and several talks from ICCE2017.
    Download PDF (1180K)
  • Yuki FURUYA, Masayoshi TAKAHASHI, Satoshi SAIKATSU, Akira YASUDA, Mich ...
    Session ID: MMS2017-32
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In this paper, we report on a small speaker system that can output 110 W and higher from an 9-V input without using analog circuits such as a D/A converter or a power amplifier. Using only digital processing, we build a low-power, fully digital speaker using a digital direct-driven speaker system (DDDSP) that can achieve high efficiency, low noise (by increasing the number of speaker units), and low THD+N. With this system, a high-quality and low-power-consumption speaker system can be used at home and at amusement facilities.
    Download PDF (804K)
  • Mami OKADA, Hidekazu SUZUKI
    Session ID: MMS2017-33
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Many consumer electronic devices in the home can now be connected to and controlled by a smartphone. However, the Bluetooth range is limited, and a user typically cannot communicate with home Bluetooth devices from outside the home. A technology is therefore proposed that virtually recognizes Bluetooth devices in the home and connects with the devices without a communication range limitation. The proposed method is extended to support Bluetooth Low Energy (BLE). A prototype of the extended system was implemented. Based on an experimental evaluation of the prototype system, we confirmed that the proposed technology enables a user to identify remotely located BLE devices through the Internet within an allowable communication delay range.
    Download PDF (945K)
  • Hiroaki OTSUBO, Akira NAKAMURA, Makoto ITAMI
    Session ID: MMS2017-34
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Dual-polarized MIMO and ultra-multilevel OFDM is proposed for next-generation digital terrestrial broadcasting. In the mobile reception of OFDM signal, inter-carrier interference(ICI) is generated by Doppler-spread. The reception characteristics are deteriorated by ICI, the improving scheme of reception characteristics is necessary. In the MIMO-ICI canceller with complexity reduction, iterative detection is adopted to improve reception characteristics. MIMO-ICI canceller using iterative detection can improve the reception characteristics with complexity reduction. In this paper, the reception characteristics are improved by MIMO-ICI canceller using iterative detection for dual-polarized MIMO ultra-multilevel OFDM under mobile reception.
    Download PDF (965K)
  • Shota HORIGUCHI, Daiki IKAMI, Kiyoharu AIZAWA
    Session ID: MMS2017-35
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    The extraction of useful deep features is important for many computer vision tasks. Deep features extracted from classification networks have proved to perform well in those tasks. On the other hand, end-to-end distance metric learning (DML) has been applied to train the feature extractor directly. However, many researches on DML did not make equitable comparisons to features extracted from classification networks, thus it is still unclear which training strategy is superior for learning feature representations. In this paper, by presenting objective comparisons between these two approaches under the same network architecture, we show that the softmax-based features are markedly better than DML features, especially when the dataset for training is large.
    Download PDF (967K)
  • Tatsuki INADA, Akie YAGURA, Yosuke YAMAMOTO, Saori HAMAGUCHI, Kazufumi ...
    Session ID: MMS2017-36
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In recent years, damage to agricultural products by wild animals has become serious. Installing traps is the main method of capturing animals, but humans must continually monitor the state of the trap to ensure they have been captured for certain which takes great labor. There is a system which can monitor and capture animals by remote control, however it is not realistic to continually monitor the state of the trap by mobile data terminal. Therefore, an automatic capturing system, which can count animals in a trap, has been developed. Also an evaluation experiment was conducted about the actual number of animals and the number counted automatically by using an image photographed by a capture trap.
    Download PDF (1290K)
  • Jiani HU, Toshihiko YAMASAKI, Kiyoharu AIZAWA
    Session ID: MMS2017-37
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Researches have shown that well tagged images are more likely to become popular, because it makes them more accessible to other users by adding as many tags as possible. However, it is not an easy task for users to annotate their content with tags that are capable of attracting popularity. Thus in this paper, we propose an recommendation approach which recommend tags that have high influence over popularity, rather than merely semantics and descriptive annotations. We then evaluate the proposed method and several existing tag recommendation strategies on a dataset of Flickr and make a comparison in both efficiency of popularity boosting and tag quality.
    Download PDF (1447K)
  • Seiya NIII, Hiroki TAKAHASHI
    Session ID: MMS2017-38
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Modern tablet devices with high performance and mobility have been widely used in daily life. The tablet computers are used in various environments that are a crowded commuter train, a kitchen for cooking and so on. In those environments, users cannot use their hands to operate a computer. Although it is effective to operate the computer by gaze, a precise gaze tracking device is required, such as an infrared camera or multiple cameras. This paper proposes a gaze gesture classification method with HMM even in a coarse gaze device.
    Download PDF (549K)
  • Shunki FUJITA, Kazuyuki SHUDO, Takeshi NISHIKAWA, Masaaki OHNISHI
    Session ID: MMS2017-39
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    High-performance cameras become available at low cost. Therefore, it is considered that there is a demand to shoot a large number of objects to be shot from a lot of viewpoints using multiple cameras. In order to utilize many cameras in soccer, music live etc., it is necessary to decide where to shoot for each camera and shoot. Depending on the video you want to shoot, it is necessary to coordinate control among the cameras so that multiple cameras do not shoot the same place. However, it is difficult for a human to operate a large number of cameras or to perform cooperative control. Therefore, in order to shoot many object using many cameras, the cameras should be automatically controlled. Preparing a large number of camera controls according to the movement of the object when automatically controlling the camera is inefficient because the number of control becomes enormous. Therefore, in this paper, we show the control technique to automatically control multiple cameras by camera operator giving shooting guidelines by scoring. By using this control technique, it is possible for a camera operator to automatically shoot videos desired to be shot automatically. In order to confirm the effectiveness of the proposed control technique, we show the results of the experiment conducted on the simulator created by Unity. Experiments where conducted by changing the scored method, and as a result the captured video changed, and it was confirmed that multiple cameras were shooting in cooperation.
    Download PDF (825K)
  • Masaya MITOBE, Kazuhiro YAMAGUCHI, Yuji SAKAMOTO
    Session ID: MMS2017-40
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In numerical simulation of image reconstruction from computer generated holograms, Fresnel diffraction integration using FFT is generally used in order to calculate propagation of light waves. There are disadvantages of impairing accuracy and flexibility of calculation in FFT. Therefore, we propose simulator of image reconstruction based on point light source method of computer-generated holograms using Fourier Transform Optical System. It enables to calculate images with high accuracy and flexibility. By the experiments, we confirm the method simulates reconstruction process successfully.
    Download PDF (801K)
  • Lingjie Wei, Yuji Sakamoto
    Session ID: MMS2017-41
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    It is known that it takes a long time to calculate a computer-generated hologram. We have developed a system to generate the animation of computer-generated hologram using fast computation algorithm. In this system, it can produce animation while maintaining high speed using the model made by CG modeler. In this paper, we will report the configuration of this system and the result of experiments.
    Download PDF (797K)
  • Fumihiro KOBAYASHI, Hiroki TAKAHASHI
    Session ID: MMS2017-42
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Target selection is given as a main operation of gaze input. In this paper, we propose target selection model for gaze input based on Fitts’s law. It is classified as gaze movement and decision movement. Gaze movement has the problem that is increased the movement time due to tremor of the gaze, and decision movement has the problem that is a method of using a screen button narrows the screen. We solve the problems using the proposed target selection model. First, we reduce a gaze movement’s difficulty with gaze likelihood, and shorten the target selection time by 1.13 second. Second, we propose don’t use a method of using screen button, use out of screen decision area.
    Download PDF (923K)
  • Hiroyoshi KOBAYASHI, Takumi CHIBA, Nobuyuki YAGI
    Session ID: MMS2017-43
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    In the virtual studio system, it is necessary to act watching synthesized image while interacting with CG objects. This causes a problem that sight line and acting are unnatural. Interaction methods, which do not disturb the shooting and do not make sight line and acting unnatural, are investigated. An interaction method is proposed to inform distance between CG objects and human body by vibrating the device which is attached to the body and is not be shot by the camera, depending on the distance. An experiment proves that the proposed method is applicable. It also suggests that time is better to sense distance and the appropriate time is from 0.5 to 0.75 sec before.
    Download PDF (503K)
  • Hakusyou DAN, Takahiro OGAWA, Miki HASEYAMA
    Session ID: MMS2017-44
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    This paper present estimation of users’ emotion evoked during listening to music. In our method, we focus on RUSBagging SVMs as a pattern recognition method to solve the data imbalance problem. RUSBagging SVMs is a pattern recognition method constructing several sub-datasets and perform classification realizing decision-level fusion. We expect that RUSBagging SVMs performs better classification compared to the conventional methods. In the experiments, we compare the result of RUSBagging SVMs and those of several methods.
    Download PDF (446K)
  • Pages 354-
    Published: 2017
    Released: May 12, 2021
    CONFERENCE PROCEEDINGS FREE ACCESS
    Download PDF (131K)
feedback
Top