Reports of the Technical Conference of the Institute of Image Electronics Engineers of Japan
Reports of the 250th Technical Conference of the Institute of Image Electronics Engineers of Japan
Showing 1-24 articles out of 24 articles from the selected issue
Date: March 23-24, 2010 Location: Sojo University
Tuesday, March 23
12:35-14:15 Chair: Kunio OHNO, Polytechnic University
  • Michiharu NIIMI, Katsunori TABATA, Hideki NODA
    Session ID: 09-06-01
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    This paper proposes a method for still image steganalysis using noise estimation on sparse code. Sparse coding is one of linear representation methods, and has the property that only a small number of the components of basic images are significantly non-zero. Basic images are not fixed but adaptive for input data, unlike DCT decomposition. It is proven that the estimation of independent component analysis model for sparse data is equivalent to sparse coding, which decompose input an image data into basic images and its coefficients. In general, the coefficients of basic images are distributed non-Gaussian. We assume that the change of image by embedding as Gausian noise. Because sparse code shrinkage can effectively separate Gausian distribution from non-Gaussian distribution, we apply the estimated noise by sparse code shrinkage to image steganalysis. In the experiments, we show our method outperforms previous steganalysis methods: Farid, Goljan and Wang's method.
    Download PDF (283K)
  • Hiroyuki YAMADA, Hirofumi NISHIKAWA
    Session ID: 09-06-02
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We present a geometric registration method for detecting digital watermarks from images. This method determines pixel-to-pixel correspondence between a copied image and its original image without referring to the original image. Feature points detected from the copied image and border lines determined to reside between the original image region and attached black region in the copied image are taken into account to determine the correspondence. The proposed method achieves locational accuracy of feature points within 2 pixels on condition that location error of detected feature points is up to 5 pixels in 720x480 copied images. The method reduces the length of video contents required to detect watermarks accurately by about 10 to 20 percent.
    Download PDF (925K)
  • Akihiro NAKAMURA, Jong G. PARK, Kotaro MATSUSHITA, Eiji NUNOHIRO
    Session ID: 09-06-03
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Tokyo University of Information Sciences receives MODIS (Moderate Resolution Imaging Spectroradiometer) data, one of the sensors equipped by NASA's Terra and Aqua satellites, and researches on "the sustainable development of economic and social structure dependent on the environment in eastern Asia". One of the major fields of the research using MODIS data is the analysis on change of environment. In order to improve the accuracy of the analysis, we developed a satellite data analysis system. In this report, the current system was modified to support web access, composite of multiple satellite data for same location, scheduling for multiple cluster nodes to configurate virtualization system. This paper presents the overview of satellite data analysis system, new feature and performance evaluation.
    Download PDF (471K)
  • - Comparison of Temporal and Spatial Normalization Condition of Feature Extraction -
    Ryota NAKAMURA, Shigeru AKAMATSU
    Session ID: 09-06-04
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    This paper describes a vision-based spoken word recognition system that utilizes, instead of audio signal, visual motion signal which is obtained from motion pictures during speech. Motion information on each pixel in the input time-series imagery was obtained by computation of optical flow, and feature values representing a spatial configuration of pixel-wise velocities were extracted for each frame image. Both starting and ending points of time for each spoken word were defined using the velocity feature values, and a high dimensional feature vector was obtained to indicate time variation of the velocity distribution within the period of utterance. As a preliminary performance evaluation of the proposed feature in spoken word recognition, discrimination test of five spoken words including A-RI-GA-TO-U and KO-N-NI-CHI-WA was conducted, and fairly promising results were achieved. Moreover, the result of using not only the motion around the mouth but also the motion extracted from other region, the spoken word recognition rather focus on the motion of the entire face than focus on only the motion around the mouth.
    Download PDF (425K)
14:25-15:15 Chair: Hiroshi OCHI, Nihon University
15:25-16:40 Chair: Naoki KOBAYASHI, Saitama Medical University
  • Hirotoshi INOMATA, Yoshinori INABA, Shigeru AKAMATSU
    Session ID: 09-06-06
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In order to generate a facial expression model, wire frame model is generally used from the viewpoint of limitation in terms of computational speed and quantity of work. However, a wire frame model is not adequate for building a more highly accurate model. On the other hand, some previous works investigated difference of facial features between male and female faces, by means of a morphable 3D face model obtained by applying PCA to a set of various 3D face data. We thus applied the latter approach to the task of facial expression generation.
    Download PDF (1038K)
  • Yuki MIYAZAKI, Kota AOKI, Hiroshi NAGAHASHI
    Session ID: 09-06-07
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Human behaviors are important information in conversation scenes. They are called non-verbal information. Especially the motions of a head like "nod", "tilt", and so on, have significant meanings as non-verbal information. Perspective n-Point problem is solved between 2D coordinates of a video sequence and 3D coordinates of a face model. Then, we can get rotation and translation parameters of the head motion. These parameters are used to quantify the motion. We performed an experiment of calculating them for 1798 frames and found some of parameters that can recognize a motion.
    Download PDF (1470K)
  • Tomomi HASUE, Kota AOKI, Hiroshi NAGAHASHI
    Session ID: 09-06-08
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In this paper, a new method for replacing a face region in video with another person's face is proposed. Haar-like features are used to detect a face region in each frame of the video. Then, HoG features are calculated from the detected region and categorized into one class that gives a facial direction by means of a Random Forest. The Random Forest is constructed in advance based on a lot of sample images with different directions and sizes that were generated from 3D face models of 13 peoples. The detected area is recognized its facial expression by using an Active Appearance Model. When facial key points are extracted by AAM, expression's information could be obtained.
    Download PDF (427K)
16:50-18:05 Chair: Yoshimasa KIMURA, Sojo University
wednesday, March 24
9:25-10:40 Chair: Kunio OHNO, Polytechnic University
  • Masaki UCHIDA, Yoshiki MIZUKAMI, Atsushi OSA, Hisashi KURODA, Motoo NO ...
    Session ID: 09-06-12
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We propose an index for evaluating quality of wound fiber. In order to determine the index we developed a system for measuring the distribution of the reflected light intensity on the surface of the wound fiber under fixed incident light. The reflected light intensity is obtained by processing an image which is shot by the camera settled at specific direction with rotation stage. The evaluation results using the proposed quality evaluation index obtained by the newly developed system showed high correlation with the visual inspection results done by fiber experts.
    Download PDF (417K)
  • - Image dependence and visual feature -
    Hidekazu SEKIZAWA, Masahiro SUZUKI, Kazutake UEHIRA
    Session ID: 09-06-13
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    It is rigid conditions at the same time for the noise characteristic under the low illumination environment though the miniaturization and making the color image reader an ultra-high-density are advanced. It previously proposed the image input process when the ultra-high-density and making to a low noise had been attempted getting a hint from the vision model. It reports on the image dependence and the corrective strategy together with the relation to the visual feature.
    Download PDF (356K)
  • Taro HONDA, Tatsuya IWASAKI, Yoshimasa KIMURA, Hiroyuki NISHI
    Session ID: 09-06-14
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We propose the method of clarifying the relation between the shape feature of the image and man's sense by the multiple regression analysis. The peripheral feature is used as the shape feature and man's sense is expressed by a subjective evaluation value defined by dissimilarity degree of fealing when man sees two binary images. When the multiple regression analysis is done using the explanatory variable in a difference of the peripheral feature of two images and the objective variable as a subjective evaluation value, the feature which has a great influence on the subjective evaluation value turns out by the partial regression coefficient. The experiment that the proposed method is applied to 30 butterfly images showed the result that man sees a horizontal side of butterfly's wing when man judges the difference of two butterfly images.
    Download PDF (464K)
10:50-12:05 Chair: Hidekazu SEKIZAWA, IIEEJ
  • Shunsuke MORI, Akira KUBOTA, Yoshinori HATORI
    Session ID: 09-06-15
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    The hybrid coding method to combine the predictive coding with the orthogonal transformation and the quantization is mainly used recently. This paper describes the hybrid coding method which uses Interpolative prediction and orthogonal quantization. Interpolative prediction uses correlations between neighboring pixels, including non-causal pixels, for high prediction performance, in contrast to conventional prediction, which uses only causal pixels. The optimal quantizing scheme is used to cancel the error that expands when decoding. Using orthogonal quantization including orthogonal transformation and quantization, we aim higher compression rate than that of the method using only quantization.
    Download PDF (463K)
  • Satoshi HISANAGA, Koji OKAMURA
    Session ID: 09-06-16
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We developed an omnidirectional image synthesis method for representing indoor maps. One of the conventional methods to construct indoor maps allowed measurement only for a small space with limited distance. Another method, studied as IBR (image based rendering), did not represent the position of objects. In our approach, we use omnidirectional images, so that maps can be constructed anywhere, even in large space. We synthesize images, so that a constructed map can represent the position of the features.
    Download PDF (1221K)
  • - Case Study on the Public Vocational Education and Training -
    Yasuaki SOGA, Tatsuya TAKEBAYASHI, Kunio OHNO
    Session ID: 09-06-17
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In this paper, application of video or animation manual to vocational education and training is described. Japanese government plans to introduce "Job Card System" and organize carrier consultants who evaluate the job cards. While the introduction of paper based job card system in Japan, web based CV (carriculum vitae) by XML format has already been introduced and carried on in US. and EC. The transformation of job card to electronic format and the evolution of vocational education and training to introduce the web system is expected. Comparison between paper based information and electronic dynamic information to apply vocational training was conducted and further analyzed. Based on the analyses, an experiment to evaluate the video movie manual for vocational training to connect a connector to LAN cable was tried.. The effectiveness of movie manual was proven through the experiment by 7 trainees. The quality of the products by training with the movie manual was much better than the products through paper manual only. In addition, duration of the period for the training with the movie manual was shorter than the period of paper manual only. Based on the experiment, consideration for the introduction of video or animation manuals to vocational education and training, and their possibility to technical communication fields were made.
    Download PDF (449K)
13:10-14:25 Chair: Hirotaka NAKANO, Osaka University
  • Kiyotaka TAMADA, Shinya KITAOKA, Yoshifumi KITAMURA, Fumio KISHINO
    Session ID: 09-06-18
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In this paper, we present a novel method of generating animation based on user-specified constraints from a small amount of motion data interactively. This method effectively combines between upper- and lower-body motion data by using simplified motion graphs. So we can generate high degree of freedom animations. When we generate animations, the method searches for an animation path in the motion graphs by improving greedy algorithm to reduce the computation time. Thus the method synthesizes the realistic motions of human characters by interactively specifying constraints. We show that our method can efficiently generate animations by illustrating synthesized motions and by evaluating the generation speed and the generation accuracy of the animations. We conclude that our method is fast enough to generate animations at interactive rates, allowing us to generate infinite-length character animations with specified constraints.
    Download PDF (1236K)
  • Takahiro TSUTSUMI, Kazunori MIYATA
    Session ID: 09-06-19
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    This research proposes a method of effective material representation considering human visual characteristics. This method adapts human visual characteristics to switch three techniques, parallax occlusion mapping, bump mapping and texture mapping, corresponding to object's level of detail. As a result, the proposed method improves drawing speed between 10-20%, comparing with the performance which applies simply a parallax occlusion mapping.
    Download PDF (2145K)
  • Shuichi OJIMA
    Session ID: 09-06-20
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We propose a simple method to improve human depth sensation for 2D images without depth information. In depth perception in 2D images, a lot of human monocular depth cues are used. Here we focus attention on shading, the improvement of depth sensation is achieved by emphasizing it. Based on the idea of painting, shade is expanded and its gradient is made steep. Retinex model is used for estimation of the reflectance images. These reflectance images are modulated to enhance shading and then are reintegrated again with the light images.
    Download PDF (480K)
14:35-16:15 Chair: Fumitaka ONO, Tokyo Polytechnic University
  • Erika ASHIKAGA, Yoshiaki TANIGUCHI, Go HASEGAWA, Hirotaka NAKANO
    Session ID: 09-06-21
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    In recent years, the demand for automatically counting pedestrians in event sites, buildings, or streets has been increased. Our research group has proposed a real-time counting method of pedestrians from video sequences where the target region is crowded and pedestrians are overlapping. In our method, the video sequences are retrieved from a stationary camera, and a liner virtual gate is set at appropriate location in the video sequences. When there is a difference of pixel value between the current frame and the background image at a pixel on the virtual gate, an optical flow whose origin is the pixel is detected. Detected optical flows are clustered based on their direction, size, and location. Then, the number of pedestrians passing through virtual gate is estimated for each cluster based on the size of cluster. In this paper, we propose a detection method of optical flow as the fundamental step of our counting method. We use block matching algorithm to detect an optical flow from video sequences. By limiting the area of search to surrounding the virtual gate, the calculation amount can be decreased and detailed detection can be accomplished in real time. Through evaluations based on actual video sequences, we confirm that 89% of optical flows can be detected successfully in real time.
    Download PDF (673K)
  • Go HASEGAWA, Sayaka KURIYAMA, Hirotaka NAKANO
    Session ID: 09-06-22
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    We have proposed the real-time estimation method of pedestrian density in video sequences. The proposed method extracts moving regions from video sequences based on background difference and inter-frame difference, and estimates the number of pedestrians from the size of the extracted region, using conversion function constructed from the prior-learned statistical data. One existing problem in the proposed method is that we utilize the conversion function constructed in an ad-hoc manner. In this paper, we introduce the simple model for the relationships between the moving region size and the number of pedestrians in the region, and construct the conversion function from the model. We evaluate the effectiveness of the proposed approach by comparison the estimated values with the true values and find that the proposed method in this paper can decrease the mean squared error by up to 32%.
    Download PDF (901K)
  • Kodai KAWANE, Tsuyoshi SASAKI, Takahiro OKAMOTO, Toshimitsu OSAJIMA, T ...
    Session ID: 09-06-23
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    A low-cost and the real-time processing are the important factors for the suspicious person monitoring system by the surveillance camera. But the motion search processing has large load. So we tried to make hardware of the motion search processing algorithm KLT-Tracker for real-time processing. And we chose low-cost device Xilinx Sparten-3 as a FPGA for implementation. For implementation in low-cost device, we propose reduction of operation amount and parallel processing, and simpliciction of oeration circuit. As a result, it actualized to implement in low-cost FPGAs and run on real-time.
    Download PDF (441K)
  • Tsuyoshi SASAKI, Kodai KAWANE, Takahiro OKAMOTO, Toshimitsu OSAJIMA, t ...
    Session ID: 09-06-24
    Published: 2010
    Released: March 27, 2012
    CONFERENCE PROCEEDINGS RESTRICTED ACCESS
    Recently, smart surveillance system has been used in various scenes to reduce the surveillance cost. Real-time motion detection system is very important to detect abnormality in surveillance video. So, we aimed to realize the high efficiently motion detection system by KLT (Kanade-Lucas-Tomasi) tracker. So far, we had proposed two methods, score control by weighted mask and adaptive feature point interval. This time, we implemented these methods to realize the real-time motion detection system by proposal methods. And we changed these proposal algorithms to suitable algorithm for hardware. We proposed architectures, weighted value generation circuit, insert position calculation circuit and feature point data update circuit. As a result of synthesis, it is possible to implement to the target device about 20% resources.
    Download PDF (696K)
feedback
Top