相較於傳統的影像判識技術著重於對像元光譜值的分析，在本研究的知識庫系統則是採用物件導向的概念來設計，較接近人的辨識過程且具有遺傳性的特性，適合用於分析類別較複雜的影像。本研究的整體處理過程如可概略的分成影像分割與知識庫判釋兩部分；在影像分割的過程中可同時加入光譜與形狀的因子加以約制，原始影像的像元經過影像分割的步驟之後形成許多區塊所組成的區塊影像，每一區塊具有其光譜以及形狀上的屬性。產生區塊影像之後，再利用知識庫來進行辨識的動作，知識庫的設計亦可加入形狀、光譜、遺傳性等條件來幫助判釋，針對影像上的每個區塊計算其在每一類別之歸屬值，完成判釋的工作。本研究是採用IKONOS衛星影像為測試資料，以同一地區的航照人工判識地物分類作為檢核的依據，成果顯示約有80％的分類正確率。 The traditional methods for image interpretation and analysis are based on the spectral characteristics of single pixel. A general idea of object -oriented programming is used in the method we proposed here. Compare to the traditional methods, the object-oriented processing of image information is nearer to the human cognitive process. Furthermore, the hierarchy between objects is well suit for analysis of complex scenes. The image interpretation process consists of two steps, “image segmentation” and “knowledge-based interpretation”. The segmentation will group pixels of image into numerous regions, which form the object image. We can bind the regions of the image by using combination of spectral and spatial factors in segment process. Each region has unique spectral and spatial characteristics. After segmentation, further classification can be achieved by calculating shape, spectral and hierarchy memberships of each region that belongs to each class in the knowledgebase. The test data of this study is an IKONOS satellite image. Using manual-interpreted airplane photo at the same area as reference for accuracy assessment. The correct classified accuracy of this study is about 80 percent.
