We have researched a real-time animation method of rivers for landscape simulations. The method approximates the shape of river wave by Stokes wave and uses the bump mapping that changes normal vectors on river surfaces to express the shake of the river at high speed because a physical shape distortion requires large amount of calculation time. Moreover, We have considered a wind effect by using 1/f noise, and we confirmed real-time processing is possible up to about 19,000 polygons.

View full abstract

This paper describes a method to reconstruct 3D structures and dynamic parameters of firework from a photograph. Theoretical analyses provide dynamic equations that constrain firework loci, enabling estimation of dynamic parameters and 3D structures from geometric features on the image plane. Experiments showed feasibility of the method.

View full abstract

[in Japanese]

View full abstract

In the West painting before modern ages, paintings drawn as the retinal image was a main current, whereas after that, visual abstract paintings came to be drawn. As a cause of this alteration, there is the pursuit of realism that it is impossible to depict by the technique of the retinal image. In this background, this paper introduces a model of the visual perceptual process such as form vision, spatial vision and color vision based on the analysis of modern paintings. Besides, our purpose is to introduce a method of transformation of entered visual objects by parameters based on the model and to construct the dialogical image transformation system implemented the parametric transformation.

View full abstract

Various methods have been proposed for Inverse Halftoning (IH) of Error Diffusion (ED) Images. Among them Table Look-Up method based on every possible combination of the bi-level output of reference pixels has been considered to be most effective and quite practical since IH function can be generated by simple averaging of the histogram of original con-tone levels of the target pixels in each context, and the image quality can be controlled by the number of reference pixels. In this paper, we will optimize the reference template and propose a hybrid system, in which each context is composed of the combination of the conventional logic pattern of prioritized reference pixels and the number of black pixels among other reference pixels. The result showed that the SNR compared under the same number of contexts is found to be improved from the conventional method especially when applied to the images not included in the training data.

View full abstract

Moving car information in traffic video signals contains a lot of important information on road safety. In this paper, we propose an algorithm that extracts and tracks Feature Points of cars. The information is indispensable for measuring moving car speed, counting the number of cars, monitoring distance between two cars, classifying running directions, and monitoring congestion situation. Utilizing Harris operator for detecting edges and corners, Feature Points are extracted, followed by block-matching to track the Feature Points in successive video frames. Many cars can be tracked at the same time automatically since the information is obtained by video signal processing. As an example, this paper shows car speed measurement result.

View full abstract

The purpose of this research is to examine the possibility of leaf recognition from the outline of leaf. DP matching and degree of circularity are used for the recognition. It is confirmed that the recognition of ten sample leaves are carried out.

View full abstract

The conventional fractal dimension is scalar-valued and cannot contain sufficient image information. For example, spatial information such as the directionality and the configurative information. Overcoming these defects, in this paper, a new type of fractal feature is proposed and applied to image analysis. The feature is a generalization of box counting dimension containing directional information. The experiment verifies that the fractal feature is very useful for analyzing texture image.

View full abstract

In this paper, we propose a reversible watermarking based on the correlation of image. In our scheme, watermarks are embedded by whether keeping existent correlation or not. We prevent the image quality degradation by changing the minimum value for destroying correlation. Computer simulations show the capability of our scheme.

View full abstract

Although a lot of research has been carried out on cursive character recognition, sufficiently high recognition rate has not been obtained. We have studied on online handprinted character &ldquo ;Hiragana” recognition using vector correlation, where patterns are represented by directional data. But, directional data sometimes can not express the geometrical information. As the result, we could not the very high classification rate. In this paper, we use the Procrustes matching together with vector correlation as the classification function. We have obtained the satisfactory recognition result.

View full abstract

Biometrics authentication has the threat of impersonation where an attacker reconstructs a user's biometric information from his user template stored in a biometric system by using the vulnerability called 'estimation' . Unlike Keys or passwords that are used in traditional token or knowledge-based authentication methods, biometric information is difficult to be canceled and reenrolled. Therefore, it is indispensable to establish template protection methods which prevent an attacker from reconstructing some biometric information from user templates. Moreover, it should be possible to revocate the user templates when their leakage occurs. In this paper, we propose a watermark-based template protection method by focusing on writer recognition systems.

View full abstract

Shakuhachi notation is a tablature written in special characters unlike a staff notation. The information processing system of shakuhachi notation is introduced in this lecture. It is the multimedia system equipped with various functions needed for processing of a Shakuhachi score, such as handwriting editing, the mutual conversion among various schools, playback with presentation of the performance method, and the mutual conversion between Standard MIDI Files. In order to realize these functions, the data representation method for shakuhachi notation is newly devised. By using this system, it is possible to create a Shakuhachi score effectively and is also expected to use as a beginner's Shakuhachi self-study tool.

View full abstract

In order to sell digital images on the internet, it is important to show the outlines of images so the purchase would be promoted. In this study, we examined human cognitive properties of partial-scrambled images. The results suggested that the identification performance is dependent on the shuffling range of digital images.

View full abstract

Camerawork can be created dynamicaly by using virtual reality technology. We performed an experiment for determining the optimal camerawork that give us the desired impressions on different screensizes. The result shows distinct screensize dependency of the optimal camerawork.

View full abstract

Problems abound regarding the issue of quality control under environments surrounding digital moving image content. Under numerous projection environments with varied color-space realizing capacities including theatrical DLP, CRT and PDP monitors, organic EL, PC, and LCD monitors of mobile phones, development as well as practice of the technology that faithfully reproduces the color of the original image as creator intended, are highly anticipated. This paper proposes a test system for examining the color management service, which enables the articulate reproduction of color of the originals, regardless of device types or environmental characteristics, by employing color-related metadata of the original image, the standard image test chart, various manufacturers' device profiles, and the calibration algorithm for faithful color reproduction.

View full abstract

In viewing and listening of the video in mobile environment, it is a subject to extend time of operation under limited power resource. When the back light of a liquid crystal display was made dark and consumption of electric power was reduced, it was a fault to become very hard to see by the fall of luminosity and contrast. Even in such a case, in order to maintain the brightness of appearance, the method of adaptive tone correction according to the statistical character of video is examined, and from the result of having experimented, while verifying about a visual effect, the application possibility to power-saving of a liquid crystal display is examined.

View full abstract

Voice messages are used for the transmission of an urgent situation to the passenger of the train, when an emergency happens in the railway. However, the transmission medium of the message is insufficient for the hearing impaired only in the voice. This paper describes the method of presenting urgent information in the train based on the concept of the universal design. To make the user understand the message faster, the formatted text and the sign language animation were displayed at the same time in the information presentation.

View full abstract

[in Japanese]

View full abstract

To determine 3D point of an object from multiple view images, corresponding points of each image must be known. Authors proposed the method of automatic setting for the correspondence points of each image. In this method, new corresponding points are automatically found by affine transformation referring known corresponding points, and pattern matching along the epipolar line. In this paper, our previous method is applied to 3D reconstruction from multiple view images, and the validity is confirmed.

View full abstract

A method of tracking a person is examined by using omnidirectional sensor with a polar coordinates image. In order to extract person's region from original image background subtraction using the HSI value which is not so influenced by a shadow, is applied. Thus, person's region is extracted by using the radius and direction information on polar coordinates. Then, for person tracking migration prediction with previous frame information is performed. Here, features of each person, such as physical size and color, are checked and updated depend on person's condition. Consequently, when no occlusion is occurred, persons can be traced within a distance of 3m from camera to persons.

View full abstract

[in Japanese]

View full abstract

Volume intersection has become popular as a method for reconstructing a 3D occluded object using multi-viewpoint cameras. This method however cannot reconstruct concave objects or objects that have holes in them. In this presentation, we propose a method to overcome this weakness by combining the volume intersection method with shape from shading and light direction estimation.We confirmed that by dividing the reconstruction into concave and convex parts of object, the whole object shape is reconstructed.

View full abstract

Interactive Holographic Television of the CG object became possible by using the calculation method of the image hologram. This research aims at the real-time holographic television by using the computer generated holographic stereogram with images captured by cameras. In this paper, we examine the reconstruction of the parallax image by using the image hologram. When each 2D image is calculated, the range of the diffraction angle is changed according to the parallax. So we can see the reconstructed image in 3D. For the experimental results, a portrait images was reconstructed clearly. And when the sight position was changed right and left, the images was switched. As a result, a clear parallax image was obtained.

View full abstract

The method to discriminate anomalous image sequences for efficiently watching monitoringvideos is proposed. Considering of applying systems composed of many monitoring cameras, the methodis required which is independent of the camera setting environment and the contents of the videos. Wepropose a method that can discriminate anomalous image sequences for more efficiently utilizing securityvideos. Considering the wide popularity of security cameras, the method is independent of the camerasetting environment and the contents of the videos. We use the spatio-temporal feature obtained byextracting the areas of change from the video. To create the input for the discrimination process, wereduce the dimensionality of the data by PCA. Discrimination is based on a 1-class SVM, which is anon-supervised learning method, and its output is the degree of anomaly of the sequence. The methodis applied to videos that simulate real environments and the results show the feasibility of determininganomalous sequences from security videos.

View full abstract

In a recent decade, a novel scheme for robust object recognition has attracted attentions when continuous image streams are input such as mutual subspace method, kernel mutual subspace method. We discuss experimental method to clarify properties of such classifiers. We propose experimental method to evaluate such classifiers.

View full abstract

In this paper, we propose a seamless and dense 3D reconstruction method that uses an uncalibrated long image sequence. First, we divide a long image sequence into subsequences. Camera motion parameters are estimated by a factorization method optimized for near planar motion from each subsequence, because we can assume that camera motion will be near planar motion in short sequences. Dense 3D points are reconstructed by applying a multi view stereo technique to each subsequence using the estimated camera motion parameters. We then find the relationships between the subsequences from the overlapping frames of adjacent subsequences. Finally, seamless and dense 3D reconstruction is achieved by merging all 3D points obtained from each subsequence. Experiments conducted on a long image sequence captured by a handheld camera while walking demonstrate that the proposed method yields good results.

View full abstract

In this paper, we propose a video digesting and delivery system "ChocoparaTV". This system automatically collects video contents that exist on the Internet based on RSS which is delivered from news sites and community sites such as blog and SNS. And the collected video contents are converted into digested videos by “ChocoPara”. Next, CH-RSS (Content Highlight RSS) - an extended RSS2.0 - is generated based on the digested videos. Users can watch the digested videos with a Flash based video RSS reader "ChocoParaTV Browser" which loads the CH-RSS, plays the digested videos and navigates the original Web pages.

View full abstract

This paper describes a new video content digesting description -CH-RSS: Content Highlight RSS- based on automatic extraction of emphasized speech and an evaluation on which the description is applied to video contents on the internet. CH-RSS applies priority level to each speech segment in descending order of “degree of emphasis” and creates a new CH-RSS document written above information. When a user requests the video content of the article, the method collects the video segments corresponding emphasized speech potion referring to the CH-RSS document. We discussed availability of non-edited videos using for blog sites applied to the proposed method.

View full abstract

In recent years, BLOG based multimedia content delivery services have gotten attention. However, there are some problems in the treatment of a multimedia content as a topic with present BLOG systems. To solve those problems and to activate communities, we propose a community support system based on BLOG systems. In this proposal, users can add some annotations and references to the scenes in audio and video contents. Moreover, users are notified the situation of the discussion to contents through information on the annotation in the syndication service, and, the users can be invited to the discussion at the right time.

View full abstract

The destruction of a soft object is targeted in this research. A soft object deforms easily and has the character where it returns easily to former shape. The object is composed by tetrahedrons, and is constructed, deformed, and destroyed by using the mass-spring model. Destruction is performed by dividing the tetrahedral vertex in consideration of the easiness of implementation. Moreover, we show that more natural destruction is possible by deciding the spring constant in proportion to the length of the tetrahedral edge, especially the object is composed of many tetrahedrons.

View full abstract

This paper presents an algorithm for realistic animation of preying act of fish based on the real scene of preying. Changing the position of each individual in a shoal of fish is realized by applying different rules depending on the number of other individuals it can see within a certain distance. The predator fish are given the characteristics that they corner the prey fish under the surface of the sea. The prey fish are given the characteristics that enough depth is maintained when they form the crowd.

View full abstract

We have been developing a walk-thought environment for a restored historical city “Edo” based on 3D computer graphics techniques. Many walkers are needed to improve the reality of such a cyber restored city. However, many people walking around the city require heavy graphics computation power. For real time walk through, we adopt a billboard animation technique with trimmed texture images captured from video movies. We have measured and evaluated the rendering performance of the billboard animation technique depending on the number of the billboards and their resolution. Experimental results show that the billboard animation technique can generate walk-through images with a couple of hundreds persons walking in the cyber city “Edo” in real time.

View full abstract

In this paper, we proposed a new method. In our proposed method, the whole area is divided into the blocks composed of 2x2 pixels, and determined a total value of pixels in each block. Secondly, the value is divided into the blocks composed of the maximum pixel value. The quotient means the number of white dots. The position of the white dots is decided by the order of pixel value. The rest is saved for the next process. Then, neighbor 2x2 blocks are merged and the above-mentioned distribution process is recursively applied. Thus, the position of white dots is progressively determined, that is to say, this algorithm works as bottom-up process. In addition, we consider improvement of our proposed method. As a result, our proposed method outperforms the Multistage division quantization method(MDQ) in the gradation region.

View full abstract

We proposed a data compression method using the Schalkwijk code for the image processed by the minimized average error method. First of all, the dithered image is divided into some blocks, in this method. Secondly, the Schalkwijk code is applied to each divided block. In the preceding work, we used only 4 x 4 pixel size for dividing into the blocks. In this paper, we consider the various block sizes such as 3 x 1, 2 x 2, and 3 x 3. In this paper, we consider various block sizes to divide the image, and we report simulation results of compression ratio.

View full abstract

Watermarking Methods are classified into two types. One is a reversible watermarking method which can restore to the original image from a watermarked one, and another is an irreversible watermarking method. However, an irreversible watermarking method cannot be utilized to some sort of application system such as a medical imaging system. In this paper, we propose a new reversible water-marking method based on a digital watermarking method using wavelet transform coefficients of several bands.

View full abstract

Density pattern image of a sine wave change is a basic change of an image. In addition, half-toning is required process when we record it with a printer. In this study, half-toned sine wave density perception conditions of sine wave are obtained experimentally.

View full abstract

We have proposed a finger indexing interface for distance lectures. We have developed an algorithm to measure where on a monitor or display is pointed at by using a stereo camera pair. In this paper, we have evaluated the accuracy of the measurement algorithm. As experimental results, about 85% of finger indexing positions are correctly measured if a user stands in front of the monitor and his index finger is 1 meter far from there. When he stands in front of but at an oblique angle to the monitor, the algorithm sometimes causes errors.

View full abstract

Although a lot of research has been carried out on depth information acquisition and feature extraction from 3- D pattern, matching measures concerning 3-D pattern recognition has not been considered adequately. In this paper, we propose a new method to define a matching measure between 3-D curves. The measure is based on Procrustes metric between two 3-D curves, and the value is scale free and rotation invariant. Therefore, the measure is promising in 3-D orbit evaluation of robot arms or other 3-D Recognition problems.

View full abstract

In order to evaluate subject's impression of handwritten characters, we propose a system generating handwriting-like characters. The system automatically generates handwriting-like characters based on genetic algorithm and the characters are shown to a subject. The subject chooses favorite characters from those that were displayed on the screen of the system. Several characters selected by the subject are used as parent characters for the production of next-generations. Handwriting-like characters in which subject's impressions are strongly reflected have been generated by the system.

View full abstract

We introduced the feature vector field that each pixel has it's feature vector, and made the prototype system which can display the pixel that became the source of the feature on the screen by using the principle of feature vector field. We confirmed the system performs well as expected.

View full abstract

In this paper we present an evaluation of using 3D-Chain Codes as representation of handwritten text for transmission over data networks. The original telewriting specification included only x and y coordinates of the handwritten text to be coded using zone codes or chain codes. Most of the actual pen based input devices can sense not only x and y coordinates, they can also sense pressure. It is important to preserve the pressure data, not only because it represents a distinguishable characteristic of the writer, but also because this parameter had shown to be useful in handwritten recognition applications. Based on the variables x, y and pressure, we construct a modified 3D-chain code to reduce the size of the data and the number of messages to be transmitted.

View full abstract

To read IDs on many objects at once at low cost, it's effective to read rough 2D codes printed on them by image recognition technologies. But there is a risk of counterfeit by mosaicking the fragments of these codes. MELON CODE enables you to embed ID number in 2D pattern similar to melon skin textures. It's difficult to counterfeit MELON CODE because of the specific characteristics of the figures.

View full abstract

This paper proposes an image data coding method for the remote-support system. The remote-support system, which helps the activity of the ambulance and the users of electric wheel chars, and so on, uses the wireless communication. Since the wireless communication is unstable and narrow bandwidth, the transmitted image data shall be compressed with very low bit-rate. However, the images compressed by the traditional methods are not suitable for the remote-support, since they are often distorted the color information and lose the sharpness of edges. To overcome this problem, the new compression method proposed in this paper is developed based on BTC (Block Truncation Coding), which has the capability to preserve the edge shape of the image. Additionally, it is introduced the following techniques: new color space conversion to improve the color resolution, and new image filter to remove the block noise generated by BTC process.

View full abstract

This paper describes high-compression PDF conversion technology for mobile terminals. High-compression PDF conversion technology converts document images into small-sized PDF files without lowering the text quality. We implemented this technology into mobile terminals and evaluated converted PDF files in terms of image quality and file size.

View full abstract

Posed by the digital revolution, various issues concerning the handling of intellectual property rights originate in the differences between continuous (analog) and discrete (digital) content. These issues require research and development on systemic designs, technologies and services, and distribution models suitable for the digital age. Network distribution of digital content is referred to as digital commerce. Enlivening such markets requires the development of products, technologies and services, and systems that satisfy the need to distribute information. To this end, the Information and Society Research Division is engaged in research and development on Digital Rights Lifecycle Management (DRLM) systems that will provide incentives to publish information by automatically increasing ratings in the commercial domain of information evaluated highly in the public domain.

View full abstract

This paper focuses on the practical way of the text indexing method which does keywords extraction automatically for the digital documents of the digital books, the patent documents, the scientific documents, the editorials and so on. To activate digital contents distribution, the content to want must be found from the inside of a lot of contents. The internet bookstores show a book by the categorized classification and is providing keywords search by the title and the author. The easier key word giving becomes possible in this proposal. We propose the easier keywords extraction.

View full abstract

This paper evaluates the effectivity of the automatic keywords extraction for the digital documents such as the digital books. The metadata is important when the content is distributed on the Internet. And, a lot of ways of the automatic keywords extraction to search a digital document were investigated. However, we think that most of the keywords extraction techniques aren't always practical. Because, in those techniques, the maintenance of the dictionary is necessary and most of the extracted keywords are noun phrases. This research group is proposing the practical way of the indexing and we verify the utility.

View full abstract

To get the digital content information, to judge whether or not we should buy it is equal to getting the digital content itself. Because the digital data can be easily copied, the digital books tend to have lost the opportunity to be bought when the purchaser would like to have a chance to check the contents before paying for it. On the other hand, when "trial reading" of the digital books cannot be carried out completely, the customer cannot judge in the usefulness and may be unable to purchase. In this paper, we propose a semi-disclosure method of text contents which can give the context and can prompt for the purchase.

View full abstract