针对大视景图象具有存储信息量巨大和基本构成图形有限的特点，提出了多级库的思路，用以大量压缩视景图象信息的存储容量，该多级库由图形库和多级映射库组成；同时给出图形库、映射库的构成方法及图形与地理位置间的映射关系和基本图形的提取方法。这种数据库具有查询速度快，查询准确性高、图象存储容量压缩比率大等特点，并能将1G字节图象用不到3Mbyte的空间存储。

For these characteristics of large landscape image such as huge saving memory and limited basic graph units to compose image, in this paper, a method based on multi level database was advanced to reduce the memory of large image. The database included two parts: graph database and mapping database. The organization mode of graph database and mapping database was presented, the mapping relation between graph units and geography position in image was put forward, the extraction of graph units from multi-level database was established. The mapping database was consisted with three level data table:the base function graph table, the function unit table, image table. The based function graph which was contructed by graph units, the function unit table which was recorded the relationship of the base function graphs described a part image, the image table was contented with all function graphs with variable situation. The reflection that established the relationship of two different data tables was described by palette, this reflection was unique. One point of high class level table reflected a unit of lower level table, by this way, it would be reduced the saving memory largely. This kind of multi level database had some characteristics such as high speed and accuracy locating, less memory of large image. By this way, 1G bytes image used memory less than 3M bytes.