Yoshitomo Yaginuma

A 3D model retrieval system for cellular phones

About a decade ago, extremely expensive computers were needed to render and visualized 3D models. However, todays economical personal computers including PDAs (personal digital assistants) and even cellular phones with a fast CPU are powerful enough to visualize 3D models interactively. Based on these kinds of hardware developments, the use of 3D models is gaining wide popularity on the Internet, and the number of 3D model databases is increasing rapidly. We have developed an experimental similarity retrieval system for 3D models for cellular phones. The system helps users find similar shaped models from databases, and the system allows users to visualize 3D models from various viewpoints by rotating the 3D models interactively. Since the use of 3D models is becoming more common on various cellular phone Web sites, these kinds of similarity retrieval algorithms and systems will play an important role in the development of such Web sites.

A partial shape matching technique for 3D model retrieval systems

The use of 3D models is gaining wide popularity since they are very important for computer graphics applications. Recently, similarity search techniques for 3D models have been investigated intensively to retrieve 3D models from the Internet. The techniques extract shape descriptors from 3D models and use these descriptors for indices of databases. Since the shape descriptors can be extracted using software, it is more efficient compared with keyword-based indices. Various shape descriptors have been proposed to improve shape similarly search results in relation to invariance of rotation, scale and translation. Most similarity search techniques are suitable for comparing each individual 3D model from databases. However, our similarity search techniques can compare not only each individual 3D model, but also similar portions of 3D models. Using our technique, each 3D model is divided into a huge number of parts, and shape descriptors are extracted from these parts to compare similarities. Although there are a large number of combinations for comparing the similarities for portions of 3D models, our shape descriptor extraction technique enables fast computing for evaluating similarities.Our system automatically decomposes 3D models into several parts by comparing angles created by normal vectors of each polygonal face. The system finds narrow angles and cuts polygonal faces into parts. Once the 3D models are decomposed, the system extracts rotation invariant shape descriptors from each part, and the descriptors are compared using a histogram to evaluate similarities. In our preliminary experiments, about 1700 3D models are decomposed into about 56,000 parts. Although there is a large number of comparisons, our shape descriptor based on equivalence class makes it possible for the system to retrieve 3D models in less than average of 10 seconds. The system allows users to find similar portions of 3D models as shown in Figure 2. Also, the system can list 3D models that contain similar parts as shown in Figure 3. Our experimental search engine can be accessed from the following web site: http://www.nime.ac.jp/web3d/

A texture energy measurement technique for 3D volumetric data

Recent advancements of 3D computer graphics hardware systems have made possible the handling of 3D volumetric data, and the amount of the available data has increased for various scientific fields. This paper proposes a pattern feature extraction method for 3D volumetric data. Pattern features are important for systems which require segmentation and classification. In this paper, the Laws texture energy approach is extended so that both 2D image data and 3D volumetric data can be handled. The Laws texture energy approach is a powerful technique for describing pattern features of 2D texture images, and it has been applied to various software applications. Our extension of the Laws texture energy approach enables direct extractions of pattern features from 3D volumetric data. Although the three dimensional extension increases the number of pattern features, the pattern features are reduced by combining similar pattern features. Our simulation software program attempts to reduce the number of shape features. The program rotates each feature along the x, y and z axes. The rotated features are evaluated if they are identical to other features, and the identical features are combined together to reduce the number of pattern features. For the experiments, artificially synthesized 3D solid textures are analyzed by using the 3D extended Laws features, and solid textures are classified by linear discriminant analysis (LDA). Our preliminary experiments show that the three dimensional extension of the Laws texture energy approach successfully classifies a certain database of 3D volumetric data.

A similarity retrieval technique for textured 3D models

This paper describes a similarity retrieval technique for textured 3D models. Various kinds of research have been conducted on similarity retrievals of 3D models since the late 1990s. Although most of the retrieval techniques focus on shape similarity of the 3D models, our technique allows users to retrieve and classify 3D models based on texture pattern similarity. To test our texture similarity retrieval technique, a set of a textured 3D model database was synthesized from 3D polygonal models and 2D texture images. The database was analyzed by software programs, and texture features were extracted from each 3D model. The extracted texture features were computed based on HLAC (higher order local autocorrelation) and fractal dimensions. Often, both kinds of texture features were used for analyzing 2D texture images. However, we extended the techniques to handle three dimensional volumetric data for extracting features from textured 3D models. Our experimental web-based retrieval system successfully retrieved textured 3D models with fairly acceptable recall-precision rates. This retrieval technique which is based on texture patterns can be used in conjunction with traditional shape similarity retrieval techniques, and the technique can enhance similarity retrieval performances.

Extended 3D HLAC pattern features for solid textures

Higher order local autocorrelation (HLAC) features are used in various 2D image applications. Recent three dimensional extensions of HLAC features (3D HLAC) have made possible the analysis of solid textures including solid textures. Typically, 3D HLAC features are computed by a limited number of three dimensional mask patterns, and the order of the mask patterns is limited up to the second order. In this paper, the orders of the mask patterns are increased up to the 26th order. Since there are a large number of combinations of mask patterns, we have implemented a simulation program for finding all their combinations in a systematic manner. Also the various subsets of the 3D HLAC mask patterns are computed by limiting the displacement regions of the masks. The 3D HLAC features were computed as local features rather than global features, and the bag of features (BoF) approach was applied for comparing local features. Our experimental program extracted pattern features from solid textures by using the extended 3D HLAC mask patterns. In our preliminary experiments, databases of solid textures were synthesized from 2D images. A similarity retrieval system for the solid textures has been implemented, and its extension to the 3D HLAC features performed well in retrievals.

A 3D Model Retrieval Based on Combinations of Partial Shape Descriptors

Similarity retrieval techniques for 3D models have been intensively investigated the last few years. The purpose has been to improve precision of the similarity retrievals, and as a result various types of shape descriptors have been proposed. Several shape descriptors use the bounding box of a 3D model during a shape descriptor extraction process, and computation of the bounding box is important for accurately identifying shape descriptors. In our previous shape descriptor extraction approaches, only one bounding box was used for each 3D model. However, use of one bounding box is a very rough approximation of the shape for certain 3D models. When the bounding box becomes very sparse for certain targeted 3D models, the approach can not compute shape descriptors accurately. In this research, we have extended the shape descriptor computation technique by using a multiple number of bounding boxes. 3D models are decomposed into multiple parts, and multiple numbers of bounding boxes are used for each decomposed part. Shape descriptors are extracted from each decomposed 3D model part independently, and they are combined with weighted values based on the proportions of area size of each decomposed part. Our preliminary experiments showed that similarity retrievals results were improved for certain 3D models by using a combination of partial shape descriptors

Video annotation tool for learning job interview

In this paper, video annotation tool for learning job interview is proposed. To visualize the difference of obtained descriptions, the proposed tool uses correspondence analysis. The results of correspondence analysis are used to give feedback to learners. By the results, the learner can understand the characteristics of his/her descriptions among the others.

Digital Library Training for Elderly Students at the Open University of Japan

The Open University of Japan (OUJ) offers distance-learning programs through courses broadcast by TV and radio, in addition to face-to-face courses offered at 50 study centers nationwide. Recently, the OUJ started to implement ICT, including Web-based delivery of courses and online registration, but these options have not been fully utilized by students. This is mainly because some older students at the OUJ had little experience in using PCs and/or the Internet. To prepare students to use the Internet and maximize Web-based learning opportunities, the OUJ began offering a digital literacy training (DLT) course at each study center in October 2010 as a 12-hour intensive course, using standardized teaching materials and a common syllabus. A series of checklist surveys was conducted before and after each course to measure the learning outcomes and perceived self-efficacy of the students. Learning outcomes and student perceptions of their digital skills were significantly improved.