- 學位論文
以不變性圖像為基礎之專利搜尋與分類
Invariant Image based Patent Search and Classification
摘要
隨著科技發展日新月異與高速的技術創新，促使以高度競爭技術為基礎之產 品與服務有關專利申請與使用量亦逐年攀升，對投資人、經營者 、或各利害關係人來說，能否進行完整專利分析之專業能力及其專利檢索分類系統在知識經濟時代下之市場競爭扮演著非常重要的角色。專利申請後經專利權責機構進行專利審 查，若符合專利法規定並無與先前技術和圖式之權利和申請內容相同時，則授予其專利權並於保護時將其技術公開。近年來，已有許多基於文字分析基礎的專利檢索 與分類工具被提出，以進行專利分析並具一定程度的效能; 然而，因專利文件的文 字存在用詞、語意或刻意規避等編寫問題，限制了以文字探勘為基礎的專利分析之準確性。本研究目的為提出以圖像不變性(含縮放、平移與旋轉)為基礎，進行專利圖式分析與檢索的新方法以期改善文字檢索之不足。研究資料下載自 Derwent 線 上專利資料庫，研究結果主要分為兩個部分，首先，針對發明專利與設計專利兩大 類之專利圖示，評估本研究之圖形不變性特徵擷取方法較諸著名專利檢索系統之 效能優劣; 其次，以專業角度客觀分析本研究與傳統以關鍵字為基礎檢索方法因作法迥異造成之差異。實驗結果並透過個案討論顯示本研究方法在兩大類專利資料 相較現有系統與方法均有良好的效能表現，說明相對於傳統的基於文本的專利搜 索的優勢，所提出的方法如何用於補足現有作業方式進行專利檢索以供各項專業使用。
並列摘要
With the proliferation of patent uses in highly competitive technology-based product or service industry, the ability to conduct comprehensive patent analysis plays a very important role in maintaining a strong market position for most inventors, sponsors, and other stakeholders. Exclusive rights are granted to patent applicants for their novel inventions if no other prior arts claimed the same things based on the text and drawings in the patent documents. A great number of keyword-based patent search tools or methods have been proposed and used in practice in recent decades in conducting patent analysis with acceptable performance. However, there are deficiencies intrinsic to the ways patent documents are written and therefore limit the degree of their success. This study aims to propose a novel method based on automated invariant patent image drawing analysis in overcoming the above-mentioned known problem. The findings of the study are two folds. First, performance is measured to evaluate the important aspects of the adopted method of this research against the well-known commercial system for both design and utility patents. Second, the strengths over traditional keyword-based patent search are illustrated owing to the fundamentally different implementation approaches. The experiment results demonstrate satisfactory performance over existing system and show how the proposed method can be used to complement the traditional patent search scheme to retrieve the relevant patents for professional use.