-
Article type: Cover
Pages
Cover1-
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
-
Article type: Index
Pages
Toc1-
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
-
Minoru HIKI, Shogo MURAMATSU, Hisakazu KIKUCHI
Article type: Article
Session ID: ME2008-189
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
In this work, we evaluate the directional adaptive nonseparable interpolation filter for sub-pixel motion compensation. In the future video coding systems, developing more efficeient interpolation filter for motion compensation is impotant subject to implove coding efficiency. In this work, we develop some nonseparable interpolation filters support different frequency region, and propose a new edge adaptive nonseparable interpolation method altering the filter by edge direction. Simulation for simple edge images and coding experiments for real video are shown.
View full abstract
-
Hideki TAGUCHI, Masahiro IWAHASHI, Tetsuya KIMURA
Article type: Article
Session ID: ME2008-190
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
In this report, we propose a new video communication method for the robot vision network in which video signals are shared by "robots" and "humans". A conventional compression algorithm, such as JPEG or MPEG discards component signals insensitive to human eyes, e.g. high frequency band and tiny signals. On the contrary, the proposed method transmits minimum components necessary for recognition by "robots". In case of communications between "robots", data size to be transmitted is reduced by the proposed method. It is also possible to communicate video signals between "humans" by transmitting only the remaining components.
View full abstract
-
Shogo MURAMATSU, Atsuyuki ADACHI
Article type: Article
Session ID: ME2008-191
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
This work proposes a novel design method of 2-D non-separable linear-phase paraunitary filter banks (LPPUFB). Especially the two-degree vanishing moment constraint is derived for the lattice structure. In the 2 × 2-decomposition case, the authors have already derived the constraint and designed non-separable symmetric orthonormal wavelets with vanishing moments of degree two. This work extends the discussion to more general 2m_y × 2m_x-decomposition case and yields a method to impose the condition to the lattice structure. The constraint is imposed directly on design parameters so that both of the linear-phase and paraunitary property are structurally guaranteed. Since the structure adopted in this work is based on the 2-D separable DCT, 2-D separable GenLOT, a popular lattice structure of DCT-based LPPUFB, is regarded as a special case. The main advantage of non-separable systems is the directional capability. This work allows us to obtain directional non-separable LOT with two-degree vanishing moments. Some design examples are shown to verify the significance of the proposed method.
View full abstract
-
Tomokazu KAWAHARA, Osamu YAMAGUCHI
Article type: Article
Session ID: ME2008-192
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
In this paper, we introduce our face recognition algorithm and its application. We mainly describe the Mutual Subspace Method (MSM) and applying Whitening Transformation to MSM. Mutual Subspace Method is the method that represents one person's face variation as a subspace. The whitening transformation is the transformation that emphasizes differences between multiple persons by increasing angles between their subspaces. Our algorithm gets high recognition accuracy using these methods.
View full abstract
-
Koichi ITO, Takafumi AOKI, Hiroshi NAKAJIMA, Koji KOBAYASHI
Article type: Article
Session ID: ME2008-193
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
This paper summarizes a high-accuracy image matching technique called Phase-Only Correlation (POC). The Phase-Only Correlation is an image matching technique using the phase components in 2D Discrete Fourier Transforms (DFTs) of given images. This technique has been successfully applied to a wide range of applications, such as industrial image recognition, passive 3D measurement, biometric authentication, video signal processing, etc. In this paper, we present applications of POC to fingerprint, iris, palmprint, face and dental radiograph recognition.
View full abstract
-
Yoshihisa Ijiri, Tadashi Hyuga, Atsushi Irie
Article type: Article
Session ID: ME2008-194
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
-
Kaoru Uchida
Article type: Article
Session ID: ME2008-195
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
Security and camera functions are important and widely used capabilities of mobile phones. I will make a brief survey over biometric authentication technology used on mobile phones, and then also explain how and-blur and image processing technologies are used to enhance camera image quality.
View full abstract
-
Shinichi TANAKA, Daigoro MIYAKAWA, Toshiyuki YOSHIDA
Article type: Article
Session ID: ME2008-196
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
The authors have proposed a video coding technique which adaptively controls the video frame rate and the average bit amount per frame to maximize an estimate for the subjective video quality (MOS) under a constant-bit-rate contstraint. The technique, however, suffers from quality variation over scenes of a target video. This paper thus proposes a technique for video quality equalization over scenes by improving the lowest quality in the target video, which is realized by adaptively varying the bit rate for each scene under the assumption that the transmission bit rate is constant and that a large-size buffer is available. The technique is then applied to videos with several scenes to verify the lowest MOS improvement. The relation between the video quality improvement and the buffer size is also investigated.
View full abstract
-
[in Japanese], [in Japanese], [in Japanese]
Article type: Article
Session ID: ME2008-197
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
-
Yasuhiro INOSHIMA, Toshiyuki YOSHIDA
Article type: Article
Session ID: ME2008-198
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
There have proposed several techniques which convert real photographs into pictrial art. Most of those techniques are based on realizing artistic brushwork or simulating behavior of pigments. Among various painting styles, this paper focuses on water-color paintings, and proposes an algorithm that generates water-color-like images from real photographs by realizing typical features on water-color-like paintings with several fundamental image processing techniques. In particular, this paper considers automatic setting for several parameters required in the conversion process, and compares an automatically generated image and manually trimmed one.
View full abstract
-
Tetsuya Fujie, Kiyoaki Aikawa
Article type: Article
Session ID: ME2008-199
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
It is necessary to improve quality for various types of audio-visual contents. This report describes a method for choosing BGM for each section of movies. This method use features extracted from the picture sequences for choosing BGM. This method can contribute to efficiently producing audio-visual contents.
View full abstract
-
Kouta Yoshida, Toshiyuki Yoshida
Article type: Article
Session ID: ME2008-200
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
The authors have proposed a 3-D shape reconstruction technique for a multi-view and multi-focus image set, which integrates 3-D shapes from different viewpoints reconstructed by the Shape From Focus (SFF) technique. This paper proposes a 3-D mesh integration technique based on features of SFF. First, a reliability metric for the depth information measured by SFF, and volumatic data weighted by the reliability is generated for a target object. Then, the 3-D shaped for the target object is obtained by applying the Marching Cubes algorithm to the volumatic data. A simulation for the proposed technique shows that the introduction of the weighting actually reduces the reconstruction error.
View full abstract
-
Naoto WATANABE, Masanobu YAMAMOTO
Article type: Article
Session ID: ME2008-201
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
We propose a system which can measure an action of a whole body from one full high-definition television camera. Using a high-resolution image, it becomes possible to measure movement of both hands and on face as well as body action. Moreover, composing the measured actions produce a 3DCG animation of the whole body.
View full abstract
-
Yasunori TATSU, Toshiyuki YOSHIDA, Yutaka Nakano
Article type: Article
Session ID: ME2008-202
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
Background subtraction technique is often affected by illumination variation. To alleviate the problem, the authors have proposed a novel background subtraction technique by extending the conventional pixel-based approach to region-based one. This paper illustrates advantages of the proposed technique over the conventional one, by applying it to several test images having various shape of targets, shading, and other interference.
View full abstract
-
Masaya IKEDA, Kan OKUBO, Norio TAGAWA
Article type: Article
Session ID: ME2008-203
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
For tracking objects, the various template matching methods are usually used. However, those cannot completely cope with apparent changes of a target object in images. On the other hand, to discriminate multiple objects in still images, the label assignment based on the MAP estimation using object's features is convenient, In this study, we propose a method which enables to track multiple objects stably without explicit tracking by extending the above MAP assignment in the temporal direction.
View full abstract
-
Article type: Appendix
Pages
App1-
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
-
Article type: Appendix
Pages
App2-
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS
-
Article type: Appendix
Pages
App3-
Published: December 16, 2008
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS
FREE ACCESS