IN-DATABASE IMAGE PROCESSING IN ORACLE SPATIAL GEORASTER
Fengting Chen, Zhihai Zhang, Ivan Lucena · 2013
Geospatial images are big data. Geospatial image processing is data intensive. Oracle Spatial GeoRaster enhances the Oracle enterprise database to natively store and manage geospatial imagery, which effectively solves the data management and scalability problems. However, with the data volume growing exponentially, real-time or near realtime image processing and database query become more challenging. This paper describes the implementation strategy of the in-database image processing engine of Oracle Spatial GeoRaster and its performance benefits. First, it not only enhances the database with advanced query capabilities, such as analytical queries and queries with image aggregation or mosaicking, but also enables massive image processing inside the database. This significantly enhances the GeoRaster data management and manipulation itself. Second, performance is the key driver behind the strategy and it has three major features to provide greater performance. The first feature is it moves the image processing closer to the images instead of moving the images out of the database to the processing. This helps achieve greater performance by avoiding data movement. The second feature is parallel processing. We parallelize some of the processing to improve performance. The third feature is concurrent processing. User can leverage the power of computer clusters and the optimized load balancing to concurrently process numerous images. The GeoRaster image processing engine supports large-scale image rectification, image appending, Virtual Mosaic, and NDVI computation among others. This paper presents some performance test results. The functionalities and performance results demonstrate that this in-database image processing engine not only dramatically improves spatial query and processing capability of large-scale image databases, but also effectively solves some of the biggest performance challenges.