Language：Japanese  

CiNii Research

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1007/s10015-015-0220-2

researchmap

Language：Japanese  

J-GLOBAL

researchmap

Language：Japanese   Publisher：The Institute of Image Information and Television Engineers  

DOI： 10.3169/itej.68.J302

CiNii Research

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1587/transinf.E96.D.1882

Web of Science

researchmap

Language：Japanese  

J-GLOBAL

researchmap

Language：Japanese   Publishing type：Research paper (scientific journal)  

DOI： 10.3169/itej.66.J354

Scopus

researchmap

Language：Japanese   Publisher：The Institute of Electronics, Information and Communication Engineers  

CiNii Research

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1587/transfun.E94.A.552

Web of Science

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.2197/ipsjtcva.3.9

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.patrec.2009.06.011

Web of Science

researchmap

Language：English   Publisher：Tohoku University  

Image segmentation has recently been studied in a framework of maximum <I>a posteriori</I> estimation for the Markov random field, where the cost function representing pixel-wise likelihood and inter-pixel smoothness should be minimized. The common drawback of these studies is the decrease in performance when a foreground object and the background have similar colors. We propose the likelihood formulation in the cost function considering not only a single pixel but also its neighboring pixels, and utilizing the support vector machine to enhance the discrimination between foreground and background. The global optimal solution for our cost function can be realized by the graph cut algorithm. Experimental results show an excellent segmentation performance in many cases.

DOI： 10.4036/iis.2007.33

CiNii Research

researchmap

Other Link： https://jlc.jst.go.jp/DN/JALC/00291771207?from=CiNii

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.physa.2006.01.013

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Web of Science

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

Web of Science

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1007/s10015-004-0337-1

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1143/JPSJ.74.488

Web of Science

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

Web of Science

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1143/JPSJ.73.2790

Web of Science

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1117/12.548492

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1103/PhysRevE.66.066126

Web of Science

researchmap

Event date： 2013.12

Language：Japanese  

In this paper, we propose a method to reconstruct the color image based on joint compressed sensing. In the conventional methods, accuracy of the reconstruction is decreased since the sparsity of signal is not enough enhanced. Further more the sampling patterns are increased because of sampling the entire image. To address these problems, in the proposed method, the sparsity of signal is enhanced based on joint compressed sensing using the color correlation and local spatial similarity of image. The sampling patterns are reduced based on block compressed sensing. Therefore the high resolution color image is finally obtained from small amount of data. Experimental results show the validity of the proposed method in terms of the image quality for the sampling rate.

researchmap

Event date： 2013.6

Language：Japanese  

In recent years, the demand for Adaptive Driving Beam (ADB) technology, which controls vehicle headlights adaptively, has recently increased. Implementation of ADB requires real-time detection of the leading vehicles. In this paper, we propose novel method of detecting and tracking leading vehicles at night. The framework of our method is 1)detecting white lines to define traffic lanes, 2)searching vehicles only in the traffic lanes, and 3)tracking the captured vehicles in subsequent frames. A major benefit of our method lies in the reduction of false detection and the cutdown of processing time by restricting the search range. Experimental results quantitatively and qualitatively show the effectiveness of the proposed method.

researchmap

Event date： 2013.5

Language：Japanese  

researchmap

Event date： 2013.1

Language：English  

researchmap

Event date： 2013

researchmap

Event date： 2013

researchmap

Event date： 2013

Language：Japanese  

In this paper, we propose the novel method of detection and tracking leading vehicles. The framework of our method is 1) detecting traffic lanes, 2) searching vehicles only in the traffic lanes, and 3) tracking the captured vehicles in subsequent frames. Experimental results quantitatively and qualitatively show the effectiveness of the proposed method.

researchmap

Event date： 2013

researchmap

Event date： 2012.10

Language：Japanese  

Most of traditional methods to synthesize virtual-view images require huge amount of data storage and computational efforts. In this paper, we propose a method to reduce the amount of data and computation by using the compressed sensing. Instead of dealing with the actual images, our method performs depth estimation for compressed linear measurement vectors. Transferring these vectors to a virtual viewpoint based on the depth information, the free-view image is finally obtained through the reconstruction process. Experimental results show the validity of the proposed method in terms of the image quality and processing time for the compression rate.

researchmap

Event date： 2012

researchmap

Event date： 2012

researchmap

Event date： 2012

researchmap

Event date： 2012

Language：Japanese  

The SVM, which is used as a powerful tool for cancer/noncancer discrimination for biopsy images, has difficulty in the parameter tuning. To avoid this problem, we propose a discrimination method using a parameter-free sparse representation classifier. Experimental results show that the performance of the proposed method is comparable to that of SVM.

researchmap

Event date： 2012

Language：Japanese  

The Wavefront Coding (WFC) is a technique of extending depth of field by restoring the image blurred through the phase mask in the lens. In this paper, we aim to improve the quality of the restoration by selecting appropriate deconvolution filters according to subject-distances estimated with the stereo matching.

researchmap

Event date： 2012

Language：Japanese  

Discrimination based on the photo-response nonuniformity (PRNU) is a powerful method for source camera identification. However, strongly-shrinked images tend to degrade the performance of the algorithm. To deal with this problem, we propose a PRNU-map reconstruction method from multiple scaled images. Experimental results show the effectiveness for heavily scaled images.

researchmap

Event date： 2011.12

Language：Japanese  

In recent years, the technology called ADB (adaptive driving beam) which is the automatic control of the vehicle headlamps, has been in great demand. To control the ADB, we have aimed at detecting leading vehicles with high accuracy from vehicle camera images. In this paper, we propose gradient/color-based features and select effective ones for vehicle detection by using the AdaBoost. A detection algorithm using the selected features is also described for vehicles appearing in various sizes on images. Experimental results quantitatively and qualitatively show the high performance of the proposed method.

researchmap

Event date： 2011.12

Language：Japanese  

Disparity estimation using the multi-view point images has been studied as an important basic technology for 3D vision. To estimate disparities even for occluded and textureless regions, some methods have been proposed based on the region segmentation and the plane fitting. However, it might be difficult to collect enough learning samples for the plane fitting in oversegmented small regions. In this paper, we propose a robust method of disparity estimation to overcome this problem by using the hierarchical segmentation. Based on the segmentation results, we hierarchically implement the stereo matching with variable-shape blocks and the plane fitting. Experimental results show the effectiveness of the proposed method and provide more desirable disparity maps than the case of nonhierarchical segmentation.

researchmap

Event date： 2011.12

Language：Japanese  

researchmap

Event date： 2011.12

Language：Japanese  

researchmap

Event date： 2011

Language：English  

In the cancer detection from stained biopsy images, it is important to extract histologically discriminative characteristics. For this purpose, we propose a novel method to extract statistical and morphological features. At the first stage, we estimate cell component memberships at each pixel by applying an expectation maximization (EM) algorithm to the color information. Next we calculate the local co-occurrence of the memberships as image features. And then, linear discriminant analysis (LDA) is applied to those features for final decision of whether cancer or not, with enhancing the discrimination. In the experiments on real biopsy images of cancers, the resulting detection accuracy is superior to the other methods.

researchmap

Event date： 2011

Language：English  

This paper explains the recent trend of smart image sensors. Next, as an application of the image sensor, we introduces our 3D imaging system for the real-time synthesis of arbitrary viewpoint image by using nine CMOS image sensors we have designed and fabricated.

researchmap

Event date： 2011

Language：English  

This paper describes how to develop and analyze multi-camera system, which generates Quad-HD video in real time. We propose new high-resolution image synthesis algorithms based on disparity estimation suitable for hardware. Our experimental results showed that Quad-HD resolution was obtained by the system. We concluded that our algorithms are effective for high-resolution cameras.

researchmap

Event date： 2011

researchmap

Event date： 2010.12

Language：English  

researchmap

Event date： 2010

Language：English  

Eye-contact plays an important role for human communications in the sense that it can convey unspoken information. However, it is highly difficult to realize eye-contact in teleconferencing systems because of camera configurations. Conventional methods to overcome this difficulty mainly resorted to space-consuming optical devices such as half mirrors. In this paper, we propose an alternative approach to achieve eye-contact by techniques of arbitrary view image synthesis. In our method, multiple images captured by real cameras are converted to the virtual viewpoint (the center of the display) by homography, and evaluation of matching errors among these projected images provides the depth map and the virtual image. Furthermore, we also propose a simpler version of this method by using a single camera to save the computational costs, in which the only one real image is transformed to the virtual viewpoint based on the hypothesis that the subject is located at a predetermined distance. In this simple implementation, eye regions are separately generated by comparison with pre-captured frontal face images. Experimental results of both the methods show that the synthesized virtual images enable the eye-contact favorably.

researchmap

Event date： 2010

Language：Japanese  

本稿では,リアルタイム任意視点画像合成システムにおける,距離推定の精度改善のための方式について検討する.本システムは,5つのCMOSイメージセンサとFPGAで構成されており,センサ面から物体までの距離の推定と指定された視点からの画像の合成を行う.重み付けブロックマッチングにより,対象物体と背景の境界付近における距離推定を行う.また,画像解像度を変化させた2段階のブロックマッチングを行う手法を導入することで精度・速度の向上を図る.

researchmap

Event date： 2009

Language：English  

In arbitrary viewpoint image synthesis, methods to use shape information of subjects have been widely proposed. However, the shape estimation is generally a time-consuming task, and is not suitable for the real time implementation. In this paper, we focus on the reconstruction filtering method in which arbitrary viewpoint image synthesis is performed without the shape estimation on the assumption that subjects exist on predetermined two planes. We propose an extension of this method to a three (and more)-layered scene and derive the filters. Furthermore, the proposed method is implemented on an FPGA, and favorable image synthesis has been achieved in real time.

researchmap

Event date： 2009

Language：English  

We have studied arbitrary viewpoint image synthesis using multiple sensors based on image-based rendering, in which the depth estimation is required for generating precise virtual images. Major problems in this estimation process include the increase of computational efforts and the decrease in performance for occluded regions. To solve these difficulties, we improve our depth estimation process by using the multi-resolution method and by weighting similar pixels to the target pixel in the block matching. Furthermore, we incorporate the proposed method to our prototype hardware system using the CMOS image sensors we have originally developed. Experimental results show that our proposed method and prototype system are efficient and effective compared to the previous work and that arbitrary viewpoint images are synthesized favorably. Copyright © 2009 by the Association for Computing Machinery, Inc.

researchmap

Event date： 2009

Language：English  

researchmap

Event date： 2009

researchmap

Event date： 2008

Language：Japanese  

We have investigated the dynamic biometrics using the dimensions of irises as a feature, which have not provided the sufficient performance. In this paper, new features based on the distance from the center to edge of irises, and the Fourier descriptor are considered. Experimental results showed the validity and effectiveness of the proposed features. Further, we examine the temporal variation for the verification performance, and the robustness of the proposed method is confirmed.

researchmap

Event date： 2008

Language：Japanese  

We propose a method to compose the image of an arbitrary viewpoint from the eight cameras array. The synthesis is based on the reconstruction filter derived from the light field rendering. The image composition can be implemented in real time using an FPGA.

researchmap

Event date： 2008

Language：English  

Counting (identical) objects in images is a simple yet fundamental recognition task that requires exhaustive human effort. Automation of this task would reduce the human load significantly. In this paper, we propose a statistical method to automatically count objects in an image sequence by using Higher-order Local Auto-Correlation (HLAC) based image features and Multiple Regression Analysis (MRA). This method is based on a simple computation, which enables fast and automatic object counting in real time. We propose several methods that have different preprocessing and image features and conduct comparative experiments of counting objects (ducks in this paper) in images captured by outdoor monitoring cameras. The experimental results demonstrated the effectiveness of the proposed methods.

researchmap

Event date： 2007

Language：English  

In image matting, several algorithms that produce high-quality results have recently been proposed; however, certain parameters have to be manually determined in order to obtain favorable results, which requires significant user effort. In this paper, we propose a method of automatically evaluating alpha mattes based on either of the two criteria: mutual information or correlation coefficient between the matte and the image gradient fields, and selecting optimal parameter values. This drastically reduces the user effort required for tuning parameters in image matting. The experimental results using several matting algorithms for various images show that the automatically selected alpha mattes are similar (or the same) to the manually selected optimal ones.

researchmap

Event date： 2006

Language：English  

Image matting and segmentation, which are used to extract a foreground object from the background, are primary techniques for digital image and video editing. In digital matting, the transparency of the foreground object is considered, while segmentation performs a rigid extraction of the object. Recently, several algorithms for matting and segmentation problems have been proposed and have provided high-quality results. In this paper, we propose a unified formulation for image matting in the framework of the method of Quantification IV based on a review of the previous studies from this framework. Our method also utilizes discriminative information provided by a user (a few strokes drawn by the user). The experimental results show a favorable matting performance. © 2007 IEEE.

researchmap

Event date： 2005

Language：English  

This article discusses issues of controlling in situations where fluctuations and delay are present. A new paradigm for control schemes, termed "delayed stochastic controls", is proposed for use in such situations.

researchmap

Event date： 2004.8

Language：Japanese  

researchmap

Event date： 2004.3

Language：Japanese  

researchmap

Event date： 2004

Language：English  

researchmap

Event date： 2003.7

Language：Japanese  

Data compression is classified to two categories : lossless and lossy compression. Our preceding research provided a scheme for lossy data compression for uniformly biased Boolean messages based on perceptron. Statistical mechanics analysis showed that when the code length is infinite, the performance of this scheme can achieve the rate-distortion function, which is the theoretical limit of the performance for lossy data compression. However, in the view point of real world, it is important to evaluate the performance in finite code length. In this research, we derive the expression of error exponents which is one of guides for the performance in finite code length, employing the replica method. We also describe a construction scheme of the optimal code, which is examined by numerical experiments.

researchmap

Event date： 2003.3

Language：Japanese  

researchmap

Event date： 2002.3

Language：Japanese  

The performance of lossy data compression is investigated via methods of statistical mechanics. Data compression is classified into two categories: lossless and lossy compressions. For each category, there exists an achievable limit which represents the best compression performance. Although it is known that several practical codes asymptotically achieve the limit for lossless compression, no practical code that saturates the limit has been found for lossy compression framework yet and quest for better practical lossy compression codes is still one of the central issues in information theory. In this thesis, a lossy data compression code on the basis of input-output relations of a perceptron is proposed and its ability and limitations are analyzed from a viewpoint of statistical mechanics. Performances of two practical encoding methods based on i) a mean field approach and ii) a Markov chain Monte Carlo sampling are also examined.

researchmap

Event date： 2001.9

Language：Japanese  

researchmap

Language：Japanese   Presentation type：Poster presentation  

researchmap

Language：English   Presentation type：Poster presentation  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

Language：Japanese   Presentation type：Oral presentation (general)  

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap

researchmap