低對比生物晶片影像之瑕疵檢測

Defect Detection of Biochip with Low Contrast Image

江政隆(Cheng-Lung Chiang)

指導教授 ： 楊宏智

國立臺灣大學/工學院/機械工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201601118

摘要

生物科技產業近年來快速發展，被視為二十一世紀的重點產業之一，根據TrendForce調查預估，2020年時，生物科技產業將占台灣GDP14%，其中生物晶片全球市場有望達到260億美元。如何提升生物晶片品質成為各家廠商追求之目標。然而在品質控管上，部分廠商仍然使用人工檢測瑕疵，不但消耗人事成本，長期檢測下易造成檢測員疲勞，影響其判斷力，因此若能將自動檢測技術應用於生物晶片檢測上將能降低人事成本並提高檢測之效率及成功率。
本研究由於檢測影像對比度較低、光源也不均勻導致特徵之間邊界不明顯，對特徵分割和瑕疵檢測帶來很大的困難，因此使設定閥值時，必須考慮各區域之光源特性。本研究將提出一套演算法流程應用於低對比生物晶片之瑕疵檢測上，藉由區域性影像強化增強對比度，並結合閥值分割、邊緣偵測、霍夫轉換等方式，經改良後進行特徵分割，並用區域性二值化方法進行瑕疵檢測。該演算法成功提升檢測準確度和降低檢測時間。

關鍵字

瑕疵檢測 ； 生物晶片 ； 低對比 ； 演算法

並列摘要

In recent years, biotech industry, regarded as one of the key industries, has been developed rapidly. According to the survey made by TrendForce, biotech industry will account for 14% of GDP in Taiwan, and the market of biochip will be about 26 billion USD in 2020. Biochip technology is an important technology of biotech industry. Although many companies pursue high quality biochips, some of these companies still depend on human eyes to inspect defects, which causes lots of personnel cost and testing personnel making mistakes due to fatigue. If Automatic Optical inspection could be apply to quality control of biochips, it will not only reduce the cost but also improve the effectiveness and the accuracy of inspection.
In this research, the edges of different features are not obvious because of the contrast is very low, making it hard to set threshold to segment different features. This thesis proposes a novel algorithm for biochips with low contrast image, using several image processing techniques based on thresholding、edge detection、hough transform for image segmentation, and local thresholding for defect detection. The algorithm improves the success rate and reduces the time cost of defect detection.

並列關鍵字

defect detection ； biochip ； low contrast ； algorithm

參考文獻

3.Berkner, K., et al., Comparison of Niblack inspired binarization methods for ancient documents. 2009. 7247: p. 72470U.

5.Canny, J., A computational approach to edge detection. IEEE Transactions on pattern analysis and machine intelligence, 1986(6): p. 679-698.

6.Chia, C.-M., K.-Y. Huang, and E. Chang, Hough transform used on the spot-centroiding algorithm for the Shack–Hartmann wavefront sensor. Optical Engineering, 2016. 55(1): p. 013105-013105.

7.Chiou, Y.C., J.Z. Liu, and Y.T. Liang, Micro crack detection of multi‐crystalline silicon solar wafer using machine vision techniques. Sensor Review, 2011. 31(2): p. 154-165.

8.Duda, R.O. and P.E. Hart, Use of the Hough transformation to detect lines and curves in pictures. Communications of the ACM, 1972. 15(1): p. 11-15.