一般於瑕疵檢測上，影像灰階分佈中如無明顯的背景與物體之分佈，將會使得影像分割難度提高。因此本研究透過獨立成份分析(Independent Component Analysis, ICA)找到瑕疵區域，後續再對瑕疵區域進行分析，避免多餘背景所造成之干擾。首先以無瑕疵影像作為ICA訓練樣本，其訓練結果可用來描述無瑕疵背景影像之區域特徵，並與待測影像進行比對，藉由正常區域與瑕疵區域之空間向量距離之差異性的不同作為瑕疵判別法則。實驗影像以TFT-LCD群輝點(Micro-Dot Defect)及Mura瑕疵為例，整合取像及ICA方法發展，達成離線學習及線上檢測。

Thresholding technique which has been widely used in the detection of defects is to automatically select an optimal threshold for separating objects from the background based on their gray-level distribution. However, this method is not suitable for TFT-LCD defects detection in this research. The basic idea of ICA is first to detect the defective regions in order to avoid unnecessary background effect before thresholding. In this research, training of samples is performed by partitioning a defect-free image into non-overlapping subimages to obtain independent components (ICs) which can be described the local features of background. Detection of defects is performed by comparing the difference of ICs between background and test subimages. In this research, we tested the performance of ICA in defects detection and achieved off-line learning and on-line detection.

[1]Y. H. Tseng, "Automatic Surface Inspection for TFT-
LCD Array Panels Using Fourier Reconstruction,"
Department of Industrial Engineering and Management,
Yuan Ze University　2003.
[2]C. C. Chen, "A study of defects of LCD panel with
wavelet method," Department of Electrical Engineering,
Yuan Ze University, 2005.
[3]A. L. Amet, A. Ertuzun, and A. Ercil, "Texture defect
detection using subband domain co-occurrencematrices,"
IEEE Southwest Symposium on Image Analysis and
Interpretation, pp. 205-210, 1998.
[4]C. J. Lu, "Automatic defect inspection for patterned
TFT-LCD panel surfaces using singular values
decomposition and independent component analysis,"
Department of Industrial Engineering and Management,
Yuan Ze University, 2004.
[5]R. Jenssen and T. Eltoft, "Independent component
analysis for texture segmentation," Pattern
Recognition, vol. 10, pp. 2301-2315, 2003.
[6]C. Y. Hung, "Automatic Surface Inspection for TFT-LCD
Array Panels Using 1D Fourier Reconstruction,"
Department of Industrial Engineering and Management,
Yuan Ze University, 2004.
[7]J. W. Jhou, "Automatic Optical Inspection on Mura
defect of TFT-LCD," Institute of Manufacturing
Engineering, Cheng Kung University, 2005.
[8]C. C. Chen, "Face Recognition by Local Features Using
Independent Component Analysis," Computer Science and
Information Engineering, Cheng Kung University, 2004.
[9]A. Hyvarinen, J. Karhunen, and E. Oja, Independent
component analysis, b ed. John Wiley & Sons, New York,
2001.
[10]J. V. Stone, Independent component analysis: a
tutorial introduction. The Mit Press, London, 2004.
[11]A. Hyvarinen, "Fast and robust fixed-point algorithms
for independent component analysis," Neural
Networks, IEEE Transactions on, vol. 10, No. 3, pp.
626-634, 1999.
[12]N. Otsu, "A threshold selection method from gray
level," IEEE Transactions on Systems, Man, and
Cybernetics, vol. 9, No. 1, pp. 62-66, 1979.
[13]"New standard: Definition of measurement index (SEMU)
for luminance Mura in FPD image quality inspection,"
Semiconductor Equipment and Materials International
(SEMI) standard, vol. 4113, pp. 242-249, 2002.