標題: 多邊形骨架表達法及其在三維工件之分類應用A Skeleton Approach to Modelling 2D Polygons and its Application to the Classification of 3D Parts 作者: 涂原彰Tu, Yuan Chang巫木誠, 莊榮宏Muh-Cherng Wu, Jung-Hong Chuang工業工程與管理學系 關鍵字: 修正簡化線骨架;初步分類;環狀碼;細部分類;逆傳遞類神經網路;revised simplified skeleton;coarse calassification;ring code;refining classification;back-propagation neural network 公開日期: 1995 摘要: 本論文旨在提出一個以工件整體外形資訊為分類依據的工件分類方法 。本論文的第一個部分，提出多邊形的修正簡化線骨架表達法，做為工件 外形資訊表達的基礎。在此表達法中，不僅克服簡化線骨架表達法(Wu and Chen 1992)中，可能出現外形相異之多邊形卻具相同骨架資訊的缺點 ，同時也進一步提升了骨架和多邊形間的拓樸類似性，免除了三維方形工 件分類研究(Wu and Jen 1994)中，多邊形須先轉換為正交而導致資訊失 真的困擾。 第二部分則提出兩階段的工件分類方法。首先，將一個三 維工件以其三個二維投影視圖之近似多邊形來表示，然後依其外形輪廓之 方位轉換為代表性環狀碼，並據此來進行外形初步分類。第二個階段則以 修正簡化線骨架為基礎，將代表此一工件的三個多邊形以三個樹狀結構之 修正簡化線骨架來表示，再將此骨架轉為逆傳遞類神經網路之輸入向量， 並藉此網路進一步將初步分類的結果中，整體外形差異較大的工件細分出 來。最後綜合兩階段分類的結果，依工件相似程度形成工件族。 This thesis presents techniques for classifying workpieces using the revised simplified skeleton. The proposed revised simplified skeleton extends the firepropagation rules of the simplified skeleton (Wu and Chen 1992). The revised simplified skeleton, as a global shape descriptor, not only captures the globalshape features, including the skeletal feature and acute shapecorners, but alsogreatly simplifies skeleton's complexity. As a result, the revised simplifiedskeleton is an ideal shape descriptor for applications such as workpiece class-ification. The proposed classification technique is a two-phase procedure. In the first phase, workpieces are grouped coarsely according to the contours oftheir projections. The coarse classification is performed by similarity matchingon the representative ring code derived from each contour. In the second phase,workpieces within the same group are classified in a refinement fasion accordingto the revised simplified skeletons of their contours. In the refining classification, the derived revised simplified skeleton is converted first to a three structureand then to a vector representation. The vector representation will be read asan input by a back-propagation neural network.Based on the results of the neuralnetwork, workpieces within the same groupare further classified into families ina three-level hierarchical structure. URI: http://140.113.39.130/cdrfb3/record/nctu/#NT840030026http://hdl.handle.net/11536/60042 Appears in Collections: Thesis