Simple and fast methods for transforming color photographs into non-photorealistic images such as stained-glass and color-sketch images. We propose two types of methods decomposing images into homogeneous color regions. One type is an edge-based method and the other is a region-based approach. In the edge-based methods, an input image is tessellated into Apollonius cells by using the distance transform of extracted edges. This tessellation and its bump mapping produce stained-glass, quilting and tailing images. Similar method utilizing distance transform of edges is proposed for generating bubble images by arranging color disks on an input image. In the region-based approach, colors in an input image are quantized and smoothed with a mode filter for decomposing an input image into cells with smooth boundaries. Their boundary lines and region colors are then modified for generating stained glass and color sketch images.

View full abstract

Gene network is a network that denotes each gene as node and connects pairs of genes by edges. It composes of a large amount of gene cluster, and therefore they contain complex connected elements, which are often difficult to interpret and grasp.This report presents a technique for visualizing hierarchical gene network data, constructed by clustering and networking techniques. The technique assumes that each gene has multiple expression rate values, and they are clustered and networked according to the correlation of the expression rate values. We visualize the data by using the enhanced "HeiankyoView", which has been originally presented as a large-scale hierarchical data visualization technique. Our technique makes easier to analyze specific genes and discover genes which has interesting features.

View full abstract

We engage in Study of VIEWGLE which quickly retrieves viewing data of digital images, large-capacity images, and use an algorithm to quickly extract the similar partial image in there. In this paper, we suggest the general method using the algorithm and examine the processing time and accuracy. Our method composes 3 steps. The first step is a preprocessing for large-capacity images and stores the result. After bringing a input image, the second step extracts candidates of target parts by rough similarity estimation. Finally the third step calculates similarity values between the candidates and input images and discovers optimal target parts by searching for the maximum similarity in the response surface method. We have not guaranteed the accuracy and exactitude of the partial image retrieval yet, however the result of this paper, the technique realizes nearly quick retrieval from large-capacity images.

View full abstract

It is very important to match images and music according to their impression. Our research focuses on automatic arrangement of music according to impressions of images. As the first result of our research, this paper reports a technique to automatically arrange music by inferring the impression of input images according to their color distribution and objects, and selecting a rhythm pattern according to the impression of images. As a preprocessing, the technique first provides users sample colors, sample images, and keywords of objects, and then collects the answer of users which are selection of rhythm patterns associated from the colors, images, and keywords. The technique then leads an equation to infer the association of rhythm pattern to the colors, images, and keywords. Finally, the technique automatically calculation the association values of all prepared rhythm patterns from arbitrary input images, and provides users the music arranged by the most associated rhythm pattern.

View full abstract

It is known that lack of exercise is detrimental to health, and various exercise machines which people can use easily in a room are available. But, it was difficult to get enough effects from exercise. Because expertise was necessary to ascertain what kind of exercise should have performed and how long we should exercise. Therefore, fitness machines with automatic controls are developed so that people can exercise efficiently, and exercise programs are designed with game elements to encourage regular exercise. However, there were problems that monotonous exercise makes it difficult for people to experience a sense of accomplishment, and if the degree of difficulty of the game is not set properly, people cannot perform the appropriate amount of exercise. Then we propose the exercise game, which measures the player heart rate in real time while playing a game and the contents of the game are actively adjusted, based on heart rate. Thus allowing users to perform sufficient amounts of exercise and experience a sense of accomplishment. Finally, we design an exercise game using the proposed method and evaluate.

View full abstract

Recently, a lot of interactive emergent story contents that enjoy seeing the life-like character's reaction have been seen. In these contents, when the user interferes in the virtual world, we can see a scene that the character perceives the objects and reacts it. However, we can't see a scene that the character affects the other character's action. Then, we propose the technique for causing various chains of episodes by expressing episode that is a story segment as a hierarchical tree structure that has event's flow and the achievement condition of AND/OR. For experiment, we produce Spilant World as an interactive game that users can see the chain episodes caused by adding the objects to the virtual world and touching the objects of the virtual world. The effectiveness of our technique is evaluated by experiencing this game.

View full abstract

A particle-based method is one of the most commonly used methods for fluid simulations. It has certain advantages over other fluid simulation methods; for example, it is easy to simulate irregular and complex fluid motion with topological changes. However, the surface reconstruction for particles remains a challenging problem. In this paper, we propose a surface reconstruction method for moving particles whose motion data are given by a particle-based simulation method. Our method utilizes a level set method to reconstruct a surface because the level set method enables the surface to change its topology free according to the irregular arrangement of particles. Moving particles' data such as positions and velocities given by a gridless particle-based method are effectively used in the level set method executed on a grid space. As a result, a suitable surface is reconstructed so as to follow the moving particles with topological changes. Besides, our method makes isolated splash particles to appropriately contribute to the surface reconstruction. Some experimental results show the useful performance of our method.

View full abstract

In this paper, we introduce background, application and current environment for software development of the graphic processor (GPU). In addition, we explore possibility of the GPU computing. After the programmable shader was introduced into the GPU, the specification of GPU has been improved with the version up of shader. Although GPU has more restriction than CPU, it is able to speed up by fully programming. GPU has wide application. Especially remarkable applications are seen in the Computer Vision and the physics simulation. And it is reported that GPU is applied in the nano-technology by the distributed computing. In addition to GPU, we touch some topics of multi-core processor technologies that are innovative architectures on the latest CPU, and some usages of graphic user interfaces in the "Windows Vista" which is the one of the latest OS. Finally, we explore possibility of the future applications.

View full abstract