- 學位論文
一種基於圖像內容特徵之龜甲類甲骨拓片碎片形狀分類
A Content-Based Feature for Shape Classification of Tortoise Jiagu-Rubbings Fragments
摘要
甲骨文物在地下埋藏3000多年，經過日積月累的自然變動，如地底的壓力及水的浸潤，使這些甲骨文物還在埋葬時期就已破碎。而挖掘時的翻動，也有可能讓甲骨文物產生碎裂。考古學家針對這些碎片，盡可能地把這些碎片進行分類及綴合，以便能夠研究甲骨上的文字符號。然而，在甲骨學的研究工作中，分類及綴合是重要且極為耗時的兩個步驟，每個步驟都需消耗大量的時間。 因此本論文提出一個基於圖像內容特徵之龜甲類甲骨拓片碎片形狀分類方法，透過甲骨學家針對各類骨板定義好的形狀規則，進行拓片碎片的分類。目的為幫助甲骨學家，提供快速分類的結果與予參考。本研究主要由三個流程組成:拓片碎片前處理、碎片特徵擷取、建立資料庫和分類模型。針對《甲骨文合集》書中，23片完整龜甲片，依據九種骨板的形狀，進行切割成碎片。並利用這九種類別各23片碎片建立資料庫，分析各類別的碎片特徵，並建立分類模型。實驗顯示，本研究提出的方法，針對九種骨板碎片和非九種骨板形狀的碎片，共1926片進行分類，其分類準確度結果為95.6%，而總執行時間為561秒(總執行時間為所有輸入影像前處理、特徵擷取及碎片分類)，有效地減少甲骨學家在碎片形狀分類上所需花費的時間。
並列摘要
Many oracle bones, which were buried underground for over 3,000 years, were broken into fragments by natural causes such as subsurface stress and water infiltration. The oracle bones were also sometimes broken into smaller fragments during excavation. To facilitate research into the Jiagu characters, archaeologists classify and rejoin the fragments as much as possible. However, classifying and then rejoining fragments are considerably time-consuming steps, slowing the progress of research. This thesis proposes a content-based feature for shape classification of tortoise Jiagu rubbings fragments to classify the rubbings fragments according to the shape rules of the bone plates defined by archaeologists. The main goal is to provide archaeologists with a mode of efficient classification. This thesis consists of three parts: the preprocessing of the rubbings fragments , feature extraction, the construction of a database, and the development of a mode of classification. The work The Great Collection of the Oracle Inscriptions offers 23 complete tortoise shells. According to the nine shapes of bone plates, these 23 complete tortoise shells can each be cut into 9 pieces. Then the 23 fragments in each of the nine classes can be used in the construction of a database, analysis of the features of each class, and the design of the mode of classification. Experimental results indicate that the proposed method has 95.6% accuracy in classifying 1,926 rubbings of fragments, whether the shapes are similar to those of the nine bone plates or not. With a total execution time of 561 seconds (including all input image preprocessing, feature extraction, and classification), this method can effectively reduce the time that is required for archaeologists to classify fragments by shape.