Motion capture technology is widely used in animation and virtual reality. There are some problems with motion capture systems. We need to have special techniques to use the system. In addition, they are expensive. Therefore, the reuse of motion capture data has become widespread. To reuse motion capture data, target motion capture data must be retrieved from a huge motion capture database with a keyword search. We describe a similarity retrieval method that uses the characteristic features of motion extracted from motion capture data. The motion of human parts is extracted as individual features of motion. We focused our study on the motion of the limbs and the head. Similar motions are retrieved in steps using features at different levels of detail. Experimental results show that the method retrieves target motion capture data efficiently.

View full abstract

Normal mapping is one of the most essential technologies for creating realistic three-dimensional computer graphics. In conventional normal map compression, such as 3Dc, only x and y components are encoded and z components are restored based on the normalizing condition. We describe an intuitively comprehensive error analysis algorithm for this approach that shows the conditions under which the compression error becomes larger. Non-linear quantization algorithms based on a formula for better compression performance than the conventional approaches are also described. Experimental results using 300 normal maps demonstrate that the PSNR is improved by 0.29-0.85 dB. In addition, the quality of rendered images can be improved by 0.76-1.81 dB on average. Our algorithms are compatible with random access and highly parallel processing on GPUs.

View full abstract

We describe a method for representing continuous tone of input monochromatic images with sparse and dense random maze patterns (RMPs). The RMPs are generated by connecting the generators of centroidal Voronoi tessellations (CVTs) with minimum spanning trees (MSTs). The brightness relationship between the input image and the RMP is derived by approximating the CVT with a hexagonal lattice, and utilized to correct the brightness for the RMP. We also describe a method for generating mazes with high degrees of difficulty in which the longest solution path between two subtrees generated by deleting a branch from the MST is identified.

View full abstract

Three-dimensional models are used in a wide range of applications. The polygonal mesh representation is a well known general purpose shape model for representing 3-D information. In general, a large quantity of data is required for storage or transmission of higher resolution polygonal mesh models. Therefore, an efficient compression scheme for 3-D polygonal mesh data is needed. We describe a new geometry compression scheme for triangle mesh. Structured data is obtained by acquiring the surrounding vertices of the triangle mesh one layer at a time. Structured geometry data are obtained according to the structured vertices and compressed by multiresolutional decomposition with triangle mesh connectivity and space frequency quantization. Experimental results show that the proposed scheme is superior to conventional schemes when operating at lower bit rates.

View full abstract

We propose an efficient and effective reconstruction method for digital content to solve the digital-content shortage. First, we analyze users or producers' eye or mouse movement to examine their viewpoints on displays showing original digital content to be reconstructed. Second, characteristic features of the attention movements are extracted, and patterns of viewpoint dependence are inquired using a picture of the "attention map" and a chart of time variation in attention point movement. Finally, design support information for reconstruction and reproduction of digital content is provided. A prototype system was implemented, and several examples are given to demonstrate our proposal.

View full abstract

We have developed a color restoration method based on spectral information to achieve the multi-spectral imaging that is necessary for acquiring highly accurate color information for digital archiving. Since we need to capture and operate many images to map onto 3D objects, the system should be practical from the viewpoints of portability of the image capturing devices and operation automation. We proposed a novel method for color restoration that effectively applies accurate colors to digital still camera (DSC) images using a normalized cut technique based on spectral information. Our method enables us to obtain DSC images and spectral information in different illumination environments and synthesize high-resolution, accurate images. We have successfully used it to create digital archival contents. Unlike the methods available in related work, our method does not need a spectral database. We demonstrated our method and presented experimental results.

View full abstract

This paper presents a run-based two-scan algorithm for labeling connected components in a binary image. Un-like conventional two-scan labeling algorithms which resolve label equivalences by processing pixels one by one, our algorithm resolves label equivalences by processing runs one by one. At any time, all provisional labels that are assigned to a connected component are combined in a set, and the smallest label is used as their representative label. The corresponding relation of a provisional label and its representative label are recorded in a table. Whenever different components are found to be connected, all provisional label sets related to these connected components are merged together, and the smallest provisional label is considered to be the representative label. When the first scan is finished, all provisional labels that were assigned to each connected component in the given image will have a unique representative label. During the second scan, we need only to replace each provisional label with its representative label. Experimental results on various types of images demonstrated that our algorithm outperforms all conventional labeling algorithms.

View full abstract

We developed Mixed Reality (MR) content that reconstructs the ancient capital of Asuka-Kyo and applied a fast shading and shadowing method that uses shadowing planes. Subjective evaluation with a head-mounted display showed that viewing this content increased the audience's knowledge both of Asuka-Kyo and of MR technologies. We also conducted impression evaluation tests with and without shading and shadowing. Content with shading and shadowing received more favorable evaluations from the audience when they were asked to rate it according to how "Realistic," "Spectacular," and "Entertaining" it was.

View full abstract