條狀結構紋路檢測演算法之設計與開發

Abstract

影像處理於條紋結構紋路檢測的技術已很常見，本論文將針對條狀結構紋路影像之扭曲方式的變化，進行檢測分析是否為有影差或其他種類瑕疵。 目前隱形眼鏡影差片的檢測方式是使用特殊取像儀器對隱形眼鏡取得相關資料，再經由專家去分析判定是否為影差片，但由於分析時間過長且分析軟體太過於專業化，因此只能以抽檢的方式進行檢測。本論文開發一隱形眼鏡條狀結構紋路檢測演算法，結合特殊取像儀器所取得的影像進行檢測，來達到自動全檢的目標。 本論文利用所取得之影像中條狀結構紋路的方向性，對其使用傅立葉轉換來取得主要的紋路走向，並設計高頻譜頻帶將主要紋路遮濾掉，最後用反傅立葉轉換並對應回原始影像，將影差片及其它瑕疵片檢測出來，最後再與原取像儀器結合，進行自動化檢測，以驗證所研發演算法之適用性與有效性。

This paper developed algorithms for inspecting defects on stripes texture. The developed algorithms first used the Fourier transform to get the direction information of the stripes texture images. Then a high-spectrum bandwidth was designed to filter out the main texture. Finally, an inverse-Fourier transformation was used to find the defects on the stripes texture. The dioptric defect images of contact lens obtained as stripes texture were used as examples for evaluating the robustness and effectiveness of the developed algorithms. Experimentation results showed that the developed algo-rithms have good performance for stripes texture detection.