隱形眼鏡乾片檢測演算法之設計與開發

Abstract

近年來，軟式隱形眼鏡在成品及半成品之自動光學檢測皆已有初步之研究成果，在半成品也就是乾片的部份，其主要光學區內結構紋路之檢測演算法，尚有改善空間，本論文將針對結構紋路之扭曲度、平行度與正交度進行量化描述，發展一套快速且有效的自動檢測演算法，以判定乾片之品質。 本論文利用專業商用儀器取得乾片之影像，分析其主要光學區的結構紋路特性，利用最小平方法求得最佳配適直線，以誤差平方和(Sum of Square Errors，SSE)衡量紋路之扭曲度，並計算分析紋路的平行度與正交度等特徵值，來描述乾片主要光學區紋路的變化，以驗證本論文所提之檢測演算法之準確性和實用性。

關鍵字：隱形眼鏡、影像處理、結構紋路、瑕疵檢測

In recent years, some automated optical inspection systems for quasi-contact lens and contact lens have been developed. However, the directional textures analysis al-gorithm for optical region of quasi-contact lens should be farther improved. This thesis describes an auto-inspection algorithm for detecting defects on quasi-contact lens. Three curial features, twistity, parallelity, and orthogonality, were extracted to build effective mapping from the original features to the semantic de-scription of the directional textures. And then conveyed into control chart to deter-mine the quality of the quasi-contact lens. The dioptric defect images of quasi-contact lens obtained as directional textures were used to validate the proposed inspection algorithm. Experimentation results demonstrated that the proposed inspection algo-rithm detects the defects accurately and robustly.

Keywords：Contact lens, Image processing, Directional textures, Defect inspection