Out-of-core image techniques with extensions for WAMI

Abstract

[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT REQUEST OF AUTHOR.] Out-of-core image techniques provide for interactive visualization, analysis, and processing of extremely large images exceeding primary memory. This requires specialized data formats for fast random access to data residing on remote or secondary storage, threading architectures for improved throughput, application-level domain specific caching, and optimized rendering techniques. While these techniques have been embraced in some scientific fields such as remote sensing and digital histopathology, barriers remain to wider adoption. This work improves on existing methods with a view-centric threading architecture to improve data throughput by loosely coupling cooperating threads while reducing locking contention, and a constant-time O(1) spatially clustering cache better suited to user interaction than time-based paging of competing systems. Wider adoption is addressed with a method for fast service-less remote data access, a portable file interchange format, and a novel fast method for online transcoding of data to this out-of-core format. Specific extensions are provided for Wide Area Motion Imagery (WAMI) with spatio-temporal caching and distributed rendering for large tiled displays. A brief study is also made for extending these concepts to create a scalable 3D point cloud visualization tool using reservoir-sampled octrees and fast screen space area calculation for continuous level-of-detail. Finally, these methods are applied beyond visualization to create novel solutions to problems in WAMI including point-of-interest stabilization, dense multiview 3D reconstruction, and accurately sampled hierarchical reconstruction in city-scale photogrammetry.