觸控面板之自動化表面瑕疵檢測

Automated Surface Defect Inspection of Touch Panels

蔡環樺(Huan-Hua Tsai)

指導教授 ： 林宏達

朝陽科技大學/理工學院/工業工程與管理系/碩士(2012年)

摘要

電容式觸控面板（Capacitive touch panel, CTP）具有防水、防汙、耐刮且反應速度快等優點，促使它在各種觸控式電子產品的應用越普及化。而CTP具多層結構且存在背景紋路為線狀結構性紋路，當瑕疵面積小且發生在背景紋路上，讓檢測工作更加困難。本研究主要針對表面具結構性的方向性紋路之CTP進行自動化瑕疵檢測，利用離散傅立葉轉換（Discrete Fourier transform, DFT）與離散餘弦轉換（Discrete cosine transform, DCT）之頻譜特性，分別搭配本研究所提之米形寬帶高斯濾波（Multi-crisscross band-Gaussian filtering, MC-BGF）與三向雙寬帶高斯濾波（Three ways double band-Gaussian filtering, 3W-DBGF）方法，同時考量高能量區域之寬帶寬度與角度並兼具寬帶濾波、閥值濾波與高斯分配濾波的特性，將能量較高之頻率值刪除，經反轉回空間域後可減弱背景紋路進而達到瑕疵增強之目的，最後使用簡易統計量設定二值化閥值界限，即可準確分離瑕疵與背景。
本研究使用148張待測影像(36張無瑕疵影像與112張瑕疵影像)進行瑕疵偵測，其檢測結果顯示DFT搭配MC-BGF方法之瑕疵檢出率 為93.42%、正常區域誤判率 為1.96%、正確判斷率(CR)為98.02%；而DCT搭配3W-DBGF方法之檢測結果為92.72% 、2.98% 、97.00%(CR)，此結果說明本研究所提之方法有不錯的檢測效益。此兩種濾波方法在濾波寬帶角度小偏移、影像亮度小與中偏移時，對檢測結果影響不大。在二值化閥值參數來源選擇方面，採用訓練之正常樣本的參數設定其檢測效果較佳。因此本研究所提之方法對於環境因素變動在中度以內亦能維持該一定檢測效果，具有不錯的穩健性。

關鍵字

離散傅立葉轉換 ； 方向性紋路 ； 瑕疵檢測 ； 觸控面板 ； 離散餘弦轉換 ； 頻率域濾波

並列摘要

Capacitive touch panels (CTP) with advantages of water-proof, stain-proof, scratch-proof, fast response, are widely used in various electronic products built in touch technology functions. The surfaces of CTPs are multi-layer structured and are classified as structural textures. It is a difficult inspection task when defects embedded in surfaces of CTPs with structural textures. This research proposes the discrete Fourier transform (DFT) based multi-crisscross band-Gaussian filtering (MC-BGF) method and the discrete cosine transform (DCT) based three ways double band-Gaussian filtering (3W-DBGF) method to inspect surface defects of CTPs. The two filtering approaches design filters to filter out the frequencies of the regions of multiple high-energy bands with considering the width and angle of the high-energy band, band filtering, threshold filtering, and characteristics of filtering with Gaussian distribution. The filtered image is then transformed back to the spatial domain. In the restored image, the homogeneous line regions in the original image will have an approximately uniform gray level, whereas the defective region will be clearly retained. Finally, the restored image is segmented by a simple threshold method and defects are located.
Real testing samples (including 36 normal images and 112 defective images) randomly selected from a manufacturing process are evaluated. Experimental results show that the DFT with MC-BGF method achieves a high 93.42% flaw detection rate , a low 1.96% false alarm rate , a high 98.02% correct classification rate (CR), and the performance indices of the DCT with 3W-DBGF method are 92.72% , 2.98% , and 97.00% (CR) on directional textured surfaces of CTPs. These results indicate that the two proposed methods have good detection effects. In additions, the two filtering methods were insensitive to small shift of band filtering angle and change of image brightness. In the source selection of thresholding parameters applied to restored images for defect separation, the use of parameters from training samples is better. Therefore, the proposed methods are also to maintain good defect detection results in moderate or little changes in environmental factors and have good robustness.

並列關鍵字

Defect inspection ； Discrete cosine transform ； Frequency domain filtering ； Discrete Fourier transform ； Directional texture ； Touch panel

參考文獻

[52] 朱崇玄，「台灣電容式觸控面板產業之策略分析」，碩士論文，國立交通大學高階主管管理學程，新竹(2009)。

[1] Adamo, F., F. Attivissimo, A. D. Nisio, and M. Savino , “A low-cost inspection system for online defects assessment in satin glass,” Measurement, Vol. 42, pp.1304-1311, (2009).

[3] Begueria, S., “Validation and evaluation of predictive models in hazard assessment and risk management,” Natural Hazards, Vol. 37, pp.315-329, (2006).

[4] Bhuvanesh, A., and M. M. Ratnam, “Automatic detection of stamping defects in lead-frames using machine vision: Overcoming translational and rotational misalignment,” International Journal of Advanced Manufacturing Technology, Vol. 32, pp.1201-1210, (2006).

[5] Catalona W. J., J. P. Richie, J. B. deKernion, F. R. Ahmann, T. L. Ratliff, B. L. Dalkin, L. R. Kavoussi, M. T. MacFarlane, and P. C. Southwick, “Comparison of prostate specific antigen concentration versus prostate specific antigen density in the early detection of prostate cancer: receiver operation characteristic curves,” J Urol, vol. 152, pp.2031-2036, (1994).