| Summary: | 碩士 === 靜宜大學 === 資訊管理學系研究所 === 96 === In the image information system, it is important to develop a spatial knowledge representation in order to achieve the goal that image retrieval can be done much precisely and quickly. By spatial knowledge representation to store the spatial information, it can support spatial reasoning and similarity retrieval to help users query images to satisfy their request. RS-string and 2D Z-string are the two famous spatial knowledge representations. RS-string is based on polar coordinate system, so it can arrive at rotation-invariant. But, 2D Z-string is based on Cartesian coordinate system, its feature not only reduces the length of string by using zero-cutting, but also records the metric information that includes the size of object、location and distance of an object. Both of them overcome the drawbacks of the relative researches of 2D string. However, we find that RS-string can not attain much precisely similarity retrieval in some condition, so it may result in misjudging of similarity retrieval. Thus, in this thesis, we propose a spatial knowledge representation which is based on RS-string and we name it as RS+-string. At the same time, we overcome the difficulty to apply the concept of 2D Z-string to the polar coordinate system. In addition, RS+-string also has the advantages of RS-string and 2D Z-string, therefore it can attain more precise spatial reasoning and similarity retrieval.
|
|---|
