Research on Large Scale Image Database System

在处理遥感数据和打印图像等大规模图像时，存在着一些问题.由于一幅图像中的数据量非常大，其数据传输时间占主导地位，很容易成为瓶颈。目前通用的操作系统对于数据密集型的1/0操作来说是非常脆弱的。为了解决这个问题，通过提供多个处理器和并行处理，我们大大减少了1/0开销。在Symmetry并行处理器上进行了遥感图像传感器定标处理的大量实验。其次，为了科普其庞大的储存容量，我们开发了使用光碟系统的影像存档系统。设备驱动程序，用于写入一次设备，并集成到LTM操作系统环境中。第三，为了改善用户界面，我们探索，面向对象的数据库方法，而不是传统的基于关键字的策略。我们开发了日本Sumi图片数据库的实验系统。通过从图像中提取语义信息并将其用于检索，可以使用户有更方便的界面。对特定应用程序的特定访问方法被注册为每个对象的方法，我们发现这样的系统非常强大。我们计划继续更详细的实施。

There remains several problems to handle large imagery such as remote sensed data and printing image. Since the volume of data in one image is very large, its data transfer time is dominant and could be easily a bottle neck. Current general operating system is very fragile for data intensive 1/0 operation. To resolve this problem, by providing multiple processors and processing in parallel we reduce the 1/0 overhead dramatically. We did the extensive exeriments on Symmetry parallel processor for sensor calibration processing of remote sensed imagery. Next, in order to cope with its large capacity of storage, we developed image archiving system using optical disk system. Device driver for write once device is written and it is integrated into LTM operating system environment. Third, in order to improve the user interface, we explored,the object oriented database approach instead of conventional keyword based strategy. We developed the experimental system for Japanese Sumi picture database. By extracting the semantic information from the image and using them in retrieval, user can have more convenient interface. Particular access method to specific application is registered as a method for each object We found such system is really powerful. We plan to continue the more elaborate implementation.