Abstract
We present LiveMosaic, a multi-media mosaic platform enabling real-time image mosaic representations of multiple inputs, such as live camera feeds, video, Internet and mobile services. LiveMosaic seeks to explore our continuous exposure to the vast amounts of impressions we endure, ranging from sensory input, mass media and our ever-changing natural surroundings. The system is intended to be used in public spaces, detect movements and broadcast areas of interest as live mosaics built and selected from a multitude of static impressions, i.e., images. In addition to this, input mixtures like Internet feeds and/or mobile services are an extra option. In order to realize live mosaic representations, we have developed pre-processing techniques to parse large image data sets into multiple branches. As a result, during the mosaic creation, the system locates the best available images in the data set within milliseconds. Consequently, no specialized hardware is necessary to generate real-time image mosaics. In the following, we will look at how LiveMosaic is different in its approach to other work and further describe the system design of LiveMosaic.

Keywords: computer graphics, mosaic, non-photorealistic rendering, real-time, motion detection, augmented reality, 3D

Related work
There has been a steady increase in diversity of computationally generated, mosaic image representations. The best-known mosaic representations include static photo mosaics [Silvers et al. 1997]. Additional research has been conducted on video mosaic representations [Klein et al. 2002] in which individual tiles of the final result consist of small videos. While most mosaic generators are offline, Media Mirror [Han 2005] is a real-time mosaic system. The Media Mirror uses 200+ live tv-channels to generate a live video mosaic representation of the person standing in front of the camera input. The LiveMosaic platform is an addition to the previously mentioned systems. We have focused our research and development on real-time interactivity in public spaces and the generation of live mosaic representations that consist of, and are selected from large amounts of impressions.

System design
LiveMosaic's main dogma is to preserve the originality of individual images given that images have our artistic thought to reflect the soul. Therefore, no color corrections are applied. The platform operates autonomously and manages input, output and internal processes like motion detection, camera control, background subtraction and mosaic rendering.
We have developed pre-processing techniques to parse large image data sets into multiple branches. As a result, during the mosaic creation, the system locates the best available images in the data set within milliseconds. Consequently, no specialized hardware is necessary to generate real-time image mosaics. LiveMosaic runs on any modern desktop computer with a decent graphic card. The largest image data set we have tested consists of ~150.000 pictures and the system refreshes at a rate of 25 fps. To realize live augmented virtual effects, LiveMosaic uses the hardware acceleration available on the graphic card. Additionally, pixel- and vertex-based shaders have been implemented to enable artists to further develop plug-ins and explore LiveMosaic's potential using 3rd-party software tools of personal preference.

References
SILVERS, R., AND HAWLEY, M. 1997. Photomosaics. Henry Holt & Company, Inc. New York, New York.

KLEIN, A. W., GRANT, T., FINKELSTEIN, A., AND COHEN, M. F. 2002. Video mosaics. NPAR '02 : In Proceedings of the 2nd international Symposium on Non-Photorealistic Animation and Rendering, pp. 21-ff.

HAN, Y. 2005. Media mirror. In SIGGRAPH 2005 Sketches and Applications, 2005, ACM Press/ACM SIGGRAPH.