An animation method for creating 3D motion effects from a single realistic, fluid-like 2D image/photo is developed. Several 3D basic velocity patterns with 3D Navier Stokes equations are designed to provide a timedependent motion for a 2D image. Their velocity patterns are transformed onto a 2D image plane using perspective transformation and a 2D advection equation is used to generate a new image sequence. Two GUIs are also developed to allow quick motion generation by mouse, integrated with image-matting and image in-painting methods. For continuing textural changes, source and sink regions can be defined and used as well. Experiments with 3D visual effects using single images, such as a cloud and a waterfall, are described. Results show that our proposed image generation method and GUI have quick operability with less distorted textures.
We propose a novel geovisualization framework based on geographic data matching between maps with intersections and their corresponding intersection database on GIS. First, users select several regions to view. The GIS generates maps with intersections and registers these intersections and their contents in our system. The users then view the map content in augmented reality (AR). To identify several dozens of area maps, we propose using LLAH (Locally Likely Arrangement Hashing), a point retrieval method using local geometry of neighbor points, when retrieving the intersections. We have improved LLAH to the point where it can now retrieve intersections with a free-moving camera. Experimental results indicate that intersection retrieval-based map retrieval is possible within several dozens of maps. Additional evaluations demonstrate the robustness of our improved LLAH from several different views.
We propose a cubic 3D display named “gCubik” which allows multiple users to simultaneously view and share 3D images. The display uses an auto-stereoscopic technique. Therefore, users can see 3D objects without the need for special 3D glasses. Its cubic shape enables users to view from arbitrary angles of 360 degree by handling the cube as if the 3D objects were in a glazed-showcase. We describe an implementation approach of the gCubik. First, we investigate the necessary parameters for fabricating a cubic auto-stereoscopic display based on the integral photography (IP) technique. Next, according to these parameters, we discuss our design of a wide field of view lens array for the IP technique. Then, we explain the construction of the cubic device using the six IP displays and the application of an OpenGL-based rendering algorithm for each lens. Finally, we discuss the produced 3D images by observing them from several different viewpoints.
We propose a technique called short-term principal component analysis (ST-PCA) to analyze motion capture (MoCap) data of realistic movements in a high dimensional time series. Our ST-PCA method is successfully applied to motion beat induction, which is an important aspect in human perception. ST-PCA performs PCA in a sliding window to locally extract the major variance of the movement into the coordinates of the first principal component, thus accurately determining the desired motion beats. Our approach differs from conventional methods in that we estimate the motion beats by analyzing the motion signals as a whole rather than individually in each channel. Moreover, our algorithm is carefully designed in terms of the three characteristics of MoCap data: hierarchical structure, spatial correlation, and temporal coherence. Experimental results demonstrate that the proposed method outputs much more accurate motion beats in a wide range of motion categories, including complicated dances, than current state-of-the-art alternatives.
Synthetic focusing methods, which superimpose pre-acquired multi-view images onto post-defined surfaces, are intended to simulate a camera with a large aperture and variable focus. We developed a synthetic focusing system with increased flexibility. The flexibility comes from independent control of the optical focuses of the input multi-view image. We arranged liquid lenses in a compact 8×8 array, in which focal lengths can be changed independently using electric signals, and placed a video camera behind the lens array to capture multi-view images. We found that if the optical and synthetic focuses coincide with each other, we can obtain images with more natural blurring effects. Furthermore, we can also set the optical focuses to multiple levels to increase the target depths of the synthetic focusing.
In recent VR and entertainment fields, many kinds of large image projection systems surrounding users are used. These systems originate from the CAVE and now are evolving as multi-projector display systems. They can effectively achieve highly immersive feelings as those users feel as if they were actually in the virtual environments projected with those systems. However, a limited number of peoples can enjoy the benefit of those systems because of the huge installation space required and cost. Obviously, these facts present significant difficulty for popularization of the projection techniques. Therefore, in this paper, we propose an effective immersive projection system with just simple projectors and convex mirrors for everyday surfaces like those in a living room. We also introduce a simple calibration method for making that system easy to use for many people.
Many object surfaces are composed of layers of different physical substances, and these are known as layered surfaces. Such surfaces have more complex optical properties than diffuse surfaces and are generally incapable of being segmented. This is because their colors change with the mixture of the optical properties of the layers, which leads to the colors changing gradually instead of sharply. To tackle these problems, we focused on surfaces with two layers, and propose a novel physical model, the Spider model. Given a single input image, our goal is to segment the colors of the image on the basis of the physical model of layered surfaces and to extract the optical properties of the two layers. The end results provide us with the approximated top layer's opacities, as well as the reflection of the top and bottom layers. The latter two are equivalent to the segmented colors of both layers. We show the results of comparison with general segmentation and digital matting. Moreover, experiments with real images show that our method is effective.
We investigated effect of face perception on race and gender classification by using facial images of racially close groups such as Thai and Japanese. Participants were Japanese and Southeast Asian college students residing in Japan. The purpose of this work is to study human perception in an attempt to establish more efficient human interface systems. In a psychological experiment, the participants from Southeast Asia acquired high recognition scores of approximately 70%, which is higher than those of the Japanese participants. Since the perception rates of the Southeast Asian participants varied by the length of their stay in Japan, it can be suggested that race perception is a learnable ability. For race perception, participants focused mostly on eyes and eyebrows. These results suggest that eyes are a significant factor in race perception. For Gender perception, we found that participants focused mostly on eyes, eyebrows, and mouths.
We describe a multi-view HDTV image generation system for live sports programs. This system is compactly composed of twelve stationary HDTV cameras and three PCs. The multi-viewpoint images generated are from a user-selected gaze point. For this, we devised a projection conversion technique for projecting the selected gaze point to the center of each camera image. In addition, to generate multi-view HDTV images quickly, we developed software that enables a GPU to process the projection conversion. This system has been used for the commentary portions of live sports programs such as gymnastics since 2006.