All issues
Predecessor
33.51
Displaying 1-26 of 26 articles from this issue
- |<
- <
- 1
- >
- >|
-
Article type: Cover
Pages Cover1-
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSDownload PDF (17K) -
Article type: Index
Pages Toc1-
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSDownload PDF (129K) -
Yuta KUBOYAMA, Yoshimitsu KUROKIArticle type: Article
Session ID: AIT2009-103/ME2009-1
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSH.264/AVC is a state-of-the-art video coding standard, which has various functions to realize high compression performance. The codec prepares several modes in both intra- and inter-prediction, and chooses the best one by some criterion. Therefore, the encoder requires a heavy burden. This paper describes a fast mode decision algorithm for intra- and inter-prediction. The proposed method prunes candidates by projecting difference blocks onto the canonical basis without calculating transformed differences, and guarantees to choose the best mode. Experimental results show that the proposed method reduces computational time by 16.3% compared with the exhaustive calculation performed by Joint Model (JM) 14.0.View full abstractDownload PDF (873K) -
Yuta HIGUCHI, Yoshimitsu KUROKIArticle type: Article
Session ID: AIT2009-104/ME2009-1
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSArbitrarily shaped transform coding is a tool to achieve object-based coding. Two types of arbitrarily shaped coding, which guarantee equivalence between the number of pixels inside shapes and DCT coefficients to be coded, are proposed as [3] and [4]. These methods obtain same DCT coefficients, and the difference is that the former calculates outside pixel values from the inside whereas the latter varies the inside pixels by themselves. They are based on 1D-DCT; therefore, 1D-DCT must be executed twice, namely horizontal and vertical directions. This procedure does not guarantee the equivalence. In this paper, we extend their methods to two-dimensional transform. In addition, we discuss methods to optimize position of DCT coefficients and pixels using statistical model. We embedded the proposed method in H.264/AVC, and simulation results indicate all proposed method performed almost same in coding efficiency.View full abstractDownload PDF (901K) -
Nobuhiro FUNATSU, Yoshimitsu KUROKIArticle type: Article
Session ID: AIT2009-105/ME2009-1
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSIn data-analysis problems with a large number of dimension, principal component analysis based on L2-norm (L2-PCA) is one of the most popular methods, but L2-PCA is sensitive to outliers. Unlike L2-PCA, PCA-L1 is robust to outliers because it utilizes the L1-norm, which is less sensitive to outliers. However, PCA-L1 needs long time to calculate bases, because PCA-L1 employs an iterative algorithm to obtain each basis, and requires to calculate an eigenvector of autocorrelation matrix as an initial vector. The autocorrelation matrix needs to be recalculated for each basis. In this paper, we focus on GPGPU (General Purpose GPU). We applied the PCA-L1 algorithm to the face-recognition technology and compared the computing time of GPGPU with CPU. Simulation results show that GPGPU decreases the execution time from CPU.View full abstractDownload PDF (681K) -
Masaaki MATSUMURA, Seishi TAKAMURA, Hirohisa JOZAWAArticle type: Article
Session ID: AIT2009-106/ME2009-1
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSIn this paper, we propose multi-directional intra predictior for lossless image coding based on AVC/H.264 block scanning. By combining some multi-directional intra predictors, we achieve 0.5[%] on average (maximum 3.1[%]) more bitrate reduction than a lossless intra predictor based on AVC/H.264.View full abstractDownload PDF (817K) -
Masanori SANO, Mahito FUJII, Norio KATAYAMA, Shin'ichi SATOHArticle type: Article
Session ID: AIT2009-107/ME2009-1
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSThis paper describes a method of generating of a visual summary from archived news video that clearly shows a topic shared by several news items. To grasp the whole topic, recognizing each news item at a glance and the flow of the news items from several viewpoints is an essential and needed. Our method represents a news item with several representative images and pieces of text, and it presents three different viewpoints for visualizing the extracted images and text. The representative images are ones from which user can guess the content of the news item or which are relevant to the news item. In order to extract such images we introduce roles of shots in news video, i.e., by applying production rules. That is, by knowing the way of shooting and editing, shots/images whose roles are lead-in, expository and spotlighting the principal object can be extracted. Our method was applied to several news topics and confirmed its effectiveness.View full abstractDownload PDF (1378K) -
Ryu TAKAYAMA, Yuuki OKUZAWA, Mi Sun Ham, Masashi OKUDAIRAArticle type: Article
Session ID: AIT2009-108/ME2009-1
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSMany kinds of labels are recently printed on goods to show environmental information. However, most of them are difficult to understand intuitively. So in this paper, a camera phone based recognition system is proposed to give environmental information. An image of a label captured with a camera phone is transmitted to a server on which the image is recognized with SIFT. Then, the related information to the label is set back to the camera phone. Experimental results with very popular environmental labels concerning recycle information show effectiveness of the system.View full abstractDownload PDF (899K) -
Hiroshi INABA, Sei-ichiro KAMATAArticle type: Article
Session ID: AIT2009-109/ME2009-1
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSIn-vehicle cameras are widely used to obtain visual information around a car. In good weather conditions, high visibility images can be obtained by using the cameras. However they do not work usefully in bad conditions such as rain. In this paper, we propose a method for removing raindrops on a windshield and repairing their regions in an image sequence captured by a vehicle mounted monocular camera. We also discuss image restoration for car-mounted camera images in bad weather conditions.View full abstractDownload PDF (1570K) -
Takayuki HARA, Haike GUANArticle type: Article
Session ID: AIT2009-110/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSA noise reduction algorithm is proposed for color image sensors like CCD. It has been difficult to reduce color noise at high speed without losing image details. To solve this problem, the proposed method employs Epsilon filter to segment the image, and outputs values on the mean color line depending on the variance ratio of noise to texture in the segmented region. Experiments indicate that the proposed method delivers better performance than conventional methods in terms of low color noise, preserving image details and high speed processing.View full abstractDownload PDF (1453K) -
Ken-ichi TANAKAArticle type: Article
Session ID: AIT2009-111/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSRecently, there are a lot of one that the processor within the error diffusion method is installed in the color printer. Now, the colors other than white and the black might be generated in the area that becomes a grayscale image when the image that a grayscale image and the color image exist together in the marketed color printer is printed. Then, we propose the algorithm where the false colors that other than white and the black are not generated in the area that becomes a grayscale image in the image of the color image and the gray image that exists together.View full abstractDownload PDF (1346K) -
Taiga KOKUBO, Nobuhiko MUKAI, Makoto KOSUGIArticle type: Article
Session ID: AIT2009-112/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSA large amount of processing time is required to generate computer graphics (CG) model for a landscape walkthrough. A landscape walkthrough is easily realized by using the interpolation images generated from two real images by morphing. Although it is easy to generate morphing images for straight roads, morphing is difficult for corner images because two images for corner morphing are different. Therefore, in this paper, we propose a generation method of the interpolation images for corner morphing by extracting a common area where morphing can be performed and by combining it with the area included only in one of the two images.View full abstractDownload PDF (3266K) -
Shou KANNO, Makoto KOSUGI, Nobuhiko MUKAIArticle type: Article
Session ID: AIT2009-113/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSThis paper proposes a method to classify deciduous trees which lose leaves by features of an outward form, trunks and branches from a photo picture. First, each outward form is represented by the Fourier descriptor and divided into three types. Next, many kinds of features such as linearity of trunk and density of branches are extracted. Each feature is designed as a decision tree to classify objective deciduous trees and introduced successful classification.View full abstractDownload PDF (1009K) -
Masashi NAKAGAWA, Nobuhiko MUKAI, Makoto KOSUGIArticle type: Article
Session ID: AIT2009-114/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSIn this paper, We propose a fast collision detection method between multiple deformable objects including self-collision by using a graphics hardware. Our method uses a layered depth image (LDI) that is generated by depth peeling, and transform feedback. Therefore, it is possible to perform collision detection very fast because there is no need for reading back from GPU. The proposed method has been implemented on a PC with NVIDIA GeForce9800GT graphics card and applied to two objects with 10,000 triangles on an image-space resolution of 800x800. As a result, the computing time for overlapping polygons was about 25 milliseconds.View full abstractDownload PDF (1068K) -
Sayaka OKAMOTO, Hiroshi HANAIZUMI, Rica HAGIWARAArticle type: Article
Session ID: AIT2009-115/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSA multi-camera system was proposed for tracking persons using PC controlled cameras connected with network. In the system, observed area was freely expanded by connecting PC controlled cameras whose FOVs were overlapped with those of the neighboring. The overlapping area was used for relaying the person information among neighboring cameras. The person information was collected by using a server-client protocol. Each client PC extracts walking persons by using Background Difference. Validity of the proposed system was confirmed by numerical simulation.View full abstractDownload PDF (1196K) -
JIAN Chang, Kohei INOUE, Kiichi URAHAMAArticle type: Article
Session ID: AIT2009-116/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSThe cross bilateral filter has been presented for improving the denoising power of the bilateral filter (BF). The cross BF is a filter smoothing a target image for denoising with the aid of a noiseless supplementary image. Its practical use is, however, limited because it requires the registration of target and supplement images. In this paper, we present a self-cross BF for improving the denoising power of BF without use of any supplementary image. We firstly devise the present technique for the ordinary BF and then we apply it to a robust BF for removing mixed Gaussian/impulsive noises from target images.View full abstractDownload PDF (1686K) -
Tatsuya YUBA, Kiichi URAHAMAArticle type: Article
Session ID: AIT2009-117/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSAn NPR technique is presented for generating mazes by embedding a solution path into a dendritic line drawing made from a root or trunk line with successive addition of branches which are spread out according to the distance transform of intermediate denritic lines. After introducing half-toning dendritic line drawings, we produce self-similar dendritic line drawings. A solution path is automatically generated as a Hamilton path connecting inputted dots. Answer paths are embedded also into backgrounds in addition to dendritic lines. Self-similar maze is made by embedding a self-similar solution line into self-similar dendritic lines. Popular mazes with grid lines can be generated by using 4-neighbor distances instead of the basic Euclidean distance.View full abstractDownload PDF (1583K) -
Kohei INOUE, Kenji HARA, Kiichi URAHAMAArticle type: Article
Session ID: AIT2009-118/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSWe propose a method for clipping a rectangular region from an image by minimizing a weighted intersection of two color histograms which are constructed with pixels included in the inside and the outside of the rectangular region. Experimental results show that the proposed method can clip the regions of objects from images and remove the regions of backgrounds.View full abstractDownload PDF (1383K) -
Yosuke TAKADA, Kiichi URAHAMAArticle type: Article
Session ID: AIT2009-119/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSA method of super-resolution of a single image is presented for approximately minimizing image reconstruction errors. A high-resolution image is estimated with an approximate Newton method for minimizing the error between an input low-resolution image and smoothed high-resolution image with a reduced size. For smoothing the high-resolution image, we use an anisotropic bilateral filter with a spatial weight of DOG (Difference Of Gaussians) characteristics and additionally we adopt the unsharp masking for enhancing the edge-preservation property of the filter. We examine with experiments the effectiveness of spatial anisotropy of the bilateral filter and utilization of the mode filter and compare the PSNR and SSIM of reconstruction errors with previous methods.View full abstractDownload PDF (1670K) -
Jegoon RYU, ToshiHiro NISHIMURAArticle type: Article
Session ID: AIT2009-120/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSIn this paper, a novel fast Gaussian blurring method that uses Look-Up Table (LUT) is presented for creating a real time Difference of Gaussian (DoG) pyramid. The LUT for fast and accurate Gaussian blur is obtained from the values calculated from one-dimensional (1D) Gaussian elements and pixel values of image. To evaluate the method, computation time and similarity with a conventional two-dimensional (2D) Gaussian blur are measured, and compared to other blurring methods. Based on the computational results, the proposed method shows good performance than other blurring methods such as box filter, Stack-blur, Romain Guy's method, and recursive Gaussian blur. The proposed method could be effectively applied to create the DoG pyramid for feature extraction.View full abstractDownload PDF (1323K) -
Kunsu HWANG, ToshiHiro NISHIMURAArticle type: Article
Session ID: AIT2009-121/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSWe propose a new method for 1/f noise reduction in complementary metal oxide semiconductor image sensors (CMOS Image Sensors: CIS) by applying characteristics of noise distribution on time domain. The algorithm using time domain or stacking images which are focused on same pixel position are proposed to make up for the weakness of general method on edge region. Nevertheless, in the case of using of self-pixel position, some problems such as blocking effect or acquisition of images are occurred. In this study, we propose the algorithm 3D weighted filtering using multi-images through noise distribution. By using weight value on the center pixel of each 2D mask, the optimized algorithm is implemented. The purpose of proposed method is to obtain the best PSNR with minimum number of frame and minimum weight value. We have used White Gaussian noise distribution because 1/f noise has that distribution. Through proposed algorithm, we have obtained the better PSNR than other methods as a masking method, Wiener filter and self pixel based method. The algorithm is expected that is effective for other random noise likes as photon shot noise with Poisson distribution.View full abstractDownload PDF (1622K) -
Kazuhito Ninomiya, Nobuji TetsutaniArticle type: Article
Session ID: AIT2009-122/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSRecently, the chance to see the video contents increases. Therefore, the producer needs the video contents that do not get tired easily. In this research, we analyzed the factor getting tired in watching CM, using the subjective evaluation and the objective evaluation for the Gaze-line. In this experiment, each subject continuously watches the same contents 22 times and 40 times. As for the images easy to tire like "Color" content and "Amount of the character" content, the Gaze-line of subjects is found to tend to be concentrated under the result of the subjective evaluation and the objective evaluation. In addition, the experimental results for "Kind of the image (goods and service)", "Person's presence", and "Number of cutting" are described.View full abstractDownload PDF (830K) -
Shohei MATSUO, Seishi TAKAMURA, Hirohisa JOZAWAArticle type: Article
Session ID: AIT2009-123/ME2009-2
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSVideo coding standards employ inter-frame prediction with motion compensation to reduce temporal redundancy. Motion compensation of integer-pel accuracy is not effective for fine movements and the coding efficiency deteriorates. Therefore, motion compensation of quarter-pel accuracy was introduced into H.264/AVC. The one dimensional 6-tap filter is used for the interpolation. However, the values of the filter coefficients are constant regardless of the resolution and the characteristic of the input video. The coding tool called "adaptive interpolation filter" (AIF) was proposed in Video Coding Experts Group (VCEG) of ITU-T. AIF optimizes the filter coefficients on frame basis. However, when the frame is divided into multiple regions where each region has different characteristics, the coding efficiency could improve by optimizing them on region basis. Consequently, we propose the region-based adaptive interpolation filter taking into account the image locality. The simulation results showed that the bit-rate reduction under the same PSNR was about 2% compared to the conventional AIF.View full abstractDownload PDF (919K) -
Article type: Appendix
Pages App1-
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSDownload PDF (90K) -
Article type: Appendix
Pages App2-
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSDownload PDF (90K) -
Article type: Appendix
Pages App3-
Published: November 19, 2009
Released on J-STAGE: September 20, 2017
CONFERENCE PROCEEDINGS FREE ACCESSDownload PDF (90K)
- |<
- <
- 1
- >
- >|