品質檢驗在產品製造過程中經常扮演著重要的角色,它不僅會影響後半段製程的可靠度,還關係到自動化生產的效率。因此近幾年已從傳統人工目視檢測方法,提升至電腦視覺之自動化檢測系統,除了可降低因人工目視所造成之誤判率,還可以維持產品品質之穩定性。在產業中常見之檢測工件,其特徵大多為體積小,且具有背景紋路,造成檢測上之困難度。在文獻中已有不少針對紋路分析之相關影像處理方法,例如離散餘弦轉換(Discrete Cosine Transform, DCT)、離散傅立葉轉換(Discrete Fourier Transform, DFT)與離散小波轉換(Discrete Wavelet Transform, DWT)等常見之頻率域轉換法,可針對不同檢測工件之紋路特徵,發展出不同之自動化檢測系統。 本研究提出一以希爾伯特-黃轉換(Hilbert Huang Transform, HHT)為基礎之檢測瑕疵方法,檢測LED透鏡之可視瑕疵,此檢測方法是由二維經驗模態分解法(Bidimensional Empirical Mode Decompositional, BEMD)搭配希爾伯特轉換(Hilbert Transform, HT)所形成。首先使用BEMD分解出數個有效之固有模式函數(Intrinsic Mode Function, IMF)影像,並選擇前三張屬高頻至中頻部分之IMF頻率影像進行HT轉換,可得到相對應之瞬間幅值(Instantaneous Amplitude, IA)強度影像。將IMF頻率影像與IA強度影像對應相乘並累加後可有效增強影像之輪廓,並且能濾除掉影像中之瑕疵,最後和原始影像相減,得到差異影像,即能突顯瑕疵之區域。實驗結果顯示本研究所提方法針對LED透鏡之可視瑕疵其瑕疵檢出率達95.24%,正常區域誤判率為2.42%,說明HHT轉換應用在瑕疵檢測上之可行性。
In the literature, there are some image processing techniquesof frequency domains fordefect detection purposes, such as Discrete Cosine Transform (DCT), Discrete Fourier Transform (DFT),Discrete Wavelet Transform (DWT) and so on. And, these frequency domain transform methods are applied to develop the automated visual inspection systems for inspecting different types of surface defectsembedded on different texture patterns. This research proposes a Hilbert-Huang Transform (HHT) based approach to inspect visual defects of LEDlenses. This approach combines the schemes of the Bi-dimensional Empirical Mode Decomposition (BEMD) and the Hilbert Transform (HT). The BEMD decomposes a testing image into several Intrinsic Mode Function (IMF) images. The first three IMF images with high and middle-high frequency components are selected to conduct the HT transformation for producing three corresponding Instantaneous Amplitude (IA) strength images. We multiple the IMF images and their corresponding IA images and then combine the multiplication images to obtain a fused image where visual defects are removed and background textures are retained. By subtracting the fused image from the original image, the visual defects can be distinctly enhanced in the difference image. Experimental results show that the defect detection rates of LED lenses achieve up 95.24% and the false alarm rates lower to 2.42% by the proposed method. This demonstrates the feasibility of applying HHT to visual defect detection.