The precise geometric correction is necessary to use the observation data by remote sensing more effectively. This paper describes the method that acquires the remaining error that is able to trust necessary to precise geometric correction.

View full abstract

We describe two topics about NOAA AVHRR image data. First, we, to get the precise result ion geometric correction, correct a satellite internal clock error and give affine transformation on map image. Then, we publish corrected image data set by World Wide Web.

View full abstract

In this paper, we propose a method using IIR filters in order to simply realize the localization. First, HRTFs by a dummy-head are approximated using IIR filters. Next, sound localization experiment using the IIR filters is carried out. Finally, the effectiveness of this method is discussed.

View full abstract

We are studying Desk Top Program Production system for efficient program production. This paper focuses on a re-use of various information in the system. All information is converted according to newly defined general data format using XML that enables flexible use for other purpose. An information retrieval system is introduced as an application of the XML data format.

View full abstract

This paper describes a method to create TV programs automatically using TVML (TV program Making Language). TVML is a scripting language that we have developed for generating TV programs by computer. In this paper, we focus on automatic news program creation. We define XML tags for describing contents of a news program, and presentation data for creating the program by given XML file. We have also developed an automatic news program creation software based on the idea.

View full abstract

This paper describes a high-quality TVML (TV program Making Language) Player for use in broadcasting, We examine all the requirements for applying the TVML Player to actual broadcast. We improved a currently-used TVML Player based on the study, for example, improved frame rate of CG, added new motions for CG characters and improved moving picture quality. The results of the confirmed that the TVML can be used for producing on-air TV programs.

View full abstract

Camera panning is a technique that is widely used to exaggerate the movement of a subject. However, many problems must be overcome in order to produce effective streaks of background scenery. An image processing was used to convolute a background with several one dimensional functions, before the image of a subject was overlapped. The best convolution function that exaggerates the impression of movement was found under assessment of the processed images. The function of the image processing corresponded to the photographing of the moving subject by a panning camera which iris opened slowly then closed slowly during the shutter releasing period. The convolution function was studied

View full abstract

In this paper, we propose image magnification method "RBI-GPDCT". Radial basis interpolation that controls jaggy of magnificated images, and Gerchberg-Papoulis iterative algorithm with DCT that restores spatial high frequency components are combined into RBI-GPDCT. As a result, it is shown that the proposed method improves jaggy and restores high frequency components on magnified images.

View full abstract

A human image is important for our proposed high quality image communication, called "the multimedia ambiance communication". In this paper, a method of capturing multi-view human image is discussed first, and proposed capturing system is built next. This system allows to input multi channel of video sequence to the computer has only one video input channel.

View full abstract

In this paper, we extend a well known hierarchical block matching technique to the image registration method to cover rotation, scale and translation between acquired focused images for the preprocessing of the image fusion, reconstruction using multi-focus images. We then test the performance of this method using synthesized images and show that the proposed method gives good accuracy. In experiment, correcting two real acquied images by means of the proposed method, we were able to generate an all-focused images with good accuracy from them.

View full abstract

In this paper, we propose a novel method of acquiring a high resolution image by integrating multiple images taken by the non-calibrated camera. If we can use 3D information, we can compose the multiple images correctly. So we propose a method to get camera calibration information by using cross stripe type clue. Then we reconstruct images by two ways. One is to estimate the target position by stereo matching, if we limit our target to plane. The other method is to use light field information. By using these methods, we can automatically reconstruct a high resolution image from many images.

View full abstract

This paper describes an image based rendering technique from a multi-view images, which utilizes range images together and projects pixels from view images neighboring a given view direction. It is targeting to deal with complicated shaped objects considering work sharing with technique like geometric-model and texture mapping. An idea of setting information concerning projection width and occlusion for every pixel is introduced. This information is determined by projecting each pixel to adjoining view images and comparing the projection result to actual pixel value and depth in the projected area. Simulation shows that natural synthesized images can be obtained from view images set at every 15 deg. view direction interval.

View full abstract

We proposed a new image-based virtual studio, using two type of image components, three dimensional image components and environmental image components. We experimentally constructed the HDTV image-based virtual studio, and verified that it could realize flexible and realistic composite studio-set images and be applied to TV program production.

View full abstract

[in Japanese]

View full abstract

Camera works are efficient image representation techniques, which are often used in motion pictures. In computer vision, it is an important problem to extract and recognize moving objects from image Sequences. This paper describes an automatic recognition method for camera works given motion pictures.

View full abstract

We are studying a stereo camera system, which can control pan, tilt and zoom, for tracking an object automatically. We developed a new method of measuring three-dimensional coordinates by projecting real images on virtual images. When using this method, the camera parameters do not change even when the camera system controls the viewing angle and zoom. We used this method to develop a prototype of a camera system and tested it.

View full abstract

Estimating the depth of the objects is one of the fundamental problem of computer vision. In this paper estimating the depth of the objects using DCT (Discrete cosine transform) from 3 multi focus edimages is mentioned.

View full abstract

In the spatio-temporal video segmentation, an image is divided into some significant regions exploiting motion information for each pixel called optical flow. It is important to estimate optical flow with accuracy for exact segmentation. In this paper, four different gradient-based optical flow estimation method are compared, and their effectiveness are discussed.

View full abstract

This paper describes an optical flow estimation method with robust statistical techniques, the method of which can provide accurate estimates even in the presence of inpulse and Gaussian noise. The least-median-of-squares method can estimate optical flows for very noisy images.

View full abstract

This paper describes a method of computational complexity reduction for the gradient motion estimation with eigenvalue algorithm. The proposed method adopts the algorithm based on perturbation theorem. The experimental results have proved that the proposed method can reduce the computational complexity and estimate movement with the same accuracy as the conventional algorithm.

View full abstract

we have proposed Global Motion Senmentation algorithm(GMS). In order to eliminat the influence of corrupted areas and/or areas which follows different motion parameters, we used threshold type cost function in the GMS algorithm. Besides the threshold type cost function, there are several robust type cost functions. In this paper, we compare the efficiency of some robust type cost functions. The results of computer simulation shows that gently-sloping cost functions have an effect on stable motion parameter set estimation.

View full abstract

This paper deals with on effient method for identifying some model in an observed image by comparing egde information obtained for the model and observed image. The prototype method computers the Hausdorff distance between all possible relative positions of the model and the image, where the model is only allowed to translate with respect to the image. In this paper, we extend the prototype method to handle not only model's translation but also rotation and zooming.

View full abstract

In this paper, we propose the method that does the extraction of the clear defect department by the combination of the image processing method to a various defect. Making it easy to recognize defect department by this method we were able to make an effective system as automatic the front of judgement processing.

View full abstract

In the field of telemedicine using CRT color monitor, the color of photographic subject represented on the objective monitor is the significant information to examine the state of patients. Yet, at the present telemedicine system, this kind of representation of the color is not considered important. Therefore in this paperWe describe the importance of representing color and then introduce the system of the color matching for telemedicine.

View full abstract

In this paper, we propose a new feature for personal identification. This feature is based on normal vector of flat plane which approximated density structure of small area in face. We also extract effective area for identification using this feature. Classification results were acceptable.

View full abstract

We studied about an analysis of facial expressions based on feature points of facial parts, "eye", "eyebrow", and "mouth". We measured movement of these feature points during variation of facial expressions, "happiness", "anger", "surprise", "sadness" and "neutral" for comparison. As are result, we get several characteristics of these facial expressions.

View full abstract

This paper presents a detection method of several peoples movements from a series of image data. We propose a method to transform the appeared image size of human body into a real image size in order to extract the position and motion loci of several peoples in real space.

View full abstract

In this paper, the authors try to estimate human existence and his position from video monitoring images for nursing and security check. In the result, we recognized position of people, if they are not occluded.

View full abstract

A facial caricature represents the individual facial features very well. This paper presents a method to synthesize a facial caricature using the eigenspaces calculated independently for each of shapes and arrangement of facial parts. Feature in an input face image are emphasized by controlling coefficient value for each eigen vector. Experimental results show the effectiveness of the proposed method.

View full abstract

Natural and gentle reproduction of human gestures are required for CG animation films, medical analysis of human actions and so on. However, a human body is composed of several mutually dependent dynamic components. Then it is very difficult to reproduce the human actions faithfully. This paper describes a motion generation method for CG characters based on a simplified dynamic model.

View full abstract

This paper proposes a design of a new AV transmission system focused on IP network environment. The system is capable of MPEG-2 video encoding and storing as well as transmitting it through IP networks without throughput degradation even in high-delay links such as satellite or international channels. This store-and-forward operation overcoming all the other problems of IP network transmission is further fortified in a prototype system by background operation and distance administration functions.

View full abstract

Video file transmission protocol is described, which is used in a high quality video transmission system over IP network. The proposed protocol enables 1) high throughput transmission over long distance line, 2) resuming transmission from disruption of former transmission, and 3) to start the transmission of video file under construction. Experiments using a network delay emulator show that the proposed protocol can provide 3 Mbit/s throughput for communication link with 200 msec round trip delay.

View full abstract

A video evaluation scheme using two video signal in a duplicate transmission system proposed, in order to detect video failure in real time automatically in the viewpoint of realizing failure-free transmission. Comparison between image feature of duplicate links / inside each link leads to accurate failure detection and identification of the disrupted link.

View full abstract

This paper verifies the performance of the enhanced error resilient video decoding method using our proposed quick error detection schemes in the video signal domain. Quick error detection schemes enable the video decoder to detect illegal discontinuity of the video signal caused by the transmission errors. Our experimental results using MREG-4 video and ITU-T H.223 media multiplexer show that quick error detection would be effective in the real mobile multimedia transmission environment.

View full abstract

Digital watermarking has been proposed as a means of copyright protection and authentication of multimedia data, since it makes possible to identify the author, owner, distributor and authorized consumer of the data. A new method based on the block correlation of DCT coefficients is presented. The pirates cannot remove the signature easily since the amount of modification is depends on the original image. The embedded sequence can be extracted without using the original image. Experimental results demonstrate that the watermark is robust to JPEG compression.

View full abstract

We present in this paper a novel method of digital watermark using multiresolution analysis of image. In this method, we embed watermark in the multiresolution components of image by using a controlled quantization process. Experimental results show that the proposed method gives the watermarked image of better quality and is robust against JPEG coding.

View full abstract

We previously proposed the lossless coding scheme of volume data based on adaptive selection of variable-length-codes. In this paper we introduce an adaptive selection method of predictors which are designed for fourteen kinds of 3D autocorrelation models. In results, the proposed method can improve a coding rate by 0.10〜0.13bits/voxel.

View full abstract

Color image can be classified into different object areas with clustered color distribution. After the classification process, each object area is characterized by Principal Component Analysis. The a^* and b^* are predicted by the projection of L^* onto chromatic plane using the eigen vectors of covariance matrix and the mean vectors. Finally, the full color image is reproducted by combining the luminance with (a^*,b^*)

View full abstract

We formerly reported on the introduction of the matching pursuits technique into the region-oriented still image coding based on Voronoi diagrams. In this paper, we investigate a method which utilizes separable basis functions suitable for oblique texture. Simulation results show that the proposed scheme outperforms the previously reported scheme by 0.4 dB in SNR.

View full abstract

In this paper, we discuss the characteristics of vector quantization on the angle coordinates space, and explain that the method is appropriate for image compression and image processing. The method's interested characteristic which transforms random original image signal distribution into regular signal is the same process as DCT(Discrete Fourier transform). Therefore, we can design many application.

View full abstract

In this paper, we discuss the pattern vector quantization using self-similar nature. To decide a codebook of a conventional vector quantization is very difficulty. The means of pattern vector quantization is a method of making and sending the codebook from the image-self. And that method of coding is similar to one of fractal coding.

View full abstract

We previously reported on the 3D-DCT video coding scheme using a frame reordering method which determined the optimal order of frames from the viewpoint of coding efficiency. In this paper, we introduce a new classification method in which the order frames is optimized in each class. Simulation results indicate that this method provides about 1dB additional improvement in SNR over the above-mentioned scheme.

View full abstract

We previously reported on the adaptive prediction scheme based on variable-size blocks. This paper describes application of overlapped compensation techniques, which reduce the blocking effects in a reconstructed image. The proposed scheme can improve not only image quality but SNR, compared with the previously reported scheme.

View full abstract

In this paper, we propose the padding method for arbitrarily shaped image coding based on minimizing entropy and quantization error. From the viewpoint of optimizing rate-distortion function, it can be seen that our proposed method is more available at each bit rate than usual one.

View full abstract

Transcoding techniques are needed to convert MPEG-4 bitstreams into H.261-compatible data. We previously proposed a MPEG-4/H.261 transcoding method which restricts functions of the MPEG-4 encoder. However, this conversion increases the bitstream size. This paper discusses how to reduce this increase in data size.

View full abstract

MPEG is getting important techniques for video handling. Since the MPEG standards specify only bitstream syntax for a compressed video, a number of algorithms have been proposed in order to improve the picture quality. In this paper, a study of the coding efficiency improvement from the point of selection of frame/field structure at MPEG encoding is mentioned.

View full abstract

Low bit-rate video coding technology such as H.263 or MPEG-4 has been developed, and it has been implemented into Internet video communication software. Some products recently released, however, have not reached at the level due to requirements which correspond to real-time interactive communication. In this paper these requirements are explained, and solutions for real-time communication are studied.

View full abstract

[in Japanese]

View full abstract

It is currently discussed how to multichannel productions on the many international conventions, because audio format of digital broadcasting in Western countries has 6 audio channels. This paper describes the trend of multichannel audio productions by the Western broadcasting stations and views the microphone techniques of multichannel audio in musical productions.

View full abstract

This document presents the audio coding system for digital broadcasting applications in Japan. BS(broadcasting Satellite) digital broadcasting standard, which starts operation in 2000, has been established. MPEG-2 Advance Audio Coding(ACC) has been adopted as audio coding for BS digital broadcasting. And the final draft standards for DTTB(digital terrestrial television broadcasting) and DTSB(digital terrestrial sound broadcasting), including ACC audio coding also have been approved.

View full abstract