Crossdata Platform for Leveraging Big + Diverse Data
Prof. Koji ZETTSU
(National Institute of Information and Communications Technology, Japan)

Nowadays with the advent of Big Data, vast and heterogeneous information reflecting natural environments and social activities can be found everywhere on the Internet such as sensing data, social media or blogs. NICT Information Services Platform (ISP) project researches and develops an ICT platform for leveraging these big + diverse data. Cyber-Physical Sensing System (CPSenS) provides a cross-domain infrastructure to collect, integrate and analyze heterogeneous sensing data from real space information obtained from sensor devices and cyber space information extracted from social media and the Web. CPSenS aims to discover associations among various events in different fields by integrating and analyzing the collected sensing data and to increase comprehensive understanding of situations of both disaster and peacetime. Cross-Database Search Engine aims to facilitate the search of interdisciplinary and correlated datasets from large-scale, multi-domain and heterogeneous data. The search is based on spatial and temporal correlations among the data, on correlations between concepts and on citational correlations between data and documents. The search engine is useful in cases such as disaster information analysis or data centric science, by discovering information about natural phenomena and social phenomena related to a certain disaster from a variety of fields. In this talk, NICT ISP technologies are introduced with their practices.

Optimal Space Transmission Matrix for Image Recognition and Its Application
Prof. Xing CHEN
(Kanagawa Institute of Technology, Japan)

Image recognition for image retrieval means to find expected similar features between images. For warning systems, it means to find differential features between images. We developed a technique by using optimal space transmission matrices to implement an image retrieval system and a nature disaster warning system. In our method, two different spaces of the same image sets are created. The first space is created by original image data or metadata of the original images. The second space, an optimal space, is created by extracting expected features from the first space. The main idea is to create a transmission matrix mapping expected features from the first space onto the optimal space and removing noise features. Using the transmission matrix, images are mapped onto the optimized space. By calculating distances between images on the optimal space, similar features and differential features of images are obtained. In addition to image retrieval and nature disaster warning, we demonstrated that this technique can also be applied to remote monitoring system. We also created a game machine using this technique to recognize human action patterns.