Integrating image processing and classification technology into automated polarizing film defect inspection. (May 2018)

• Record Type: Journal Article
• Authors: Kuo, Chung-Feng Jeffrey; Lai, Chun-Yu; Kao, Chih-Hsiang; Chiu, Chin-Hsun
• Is Part Of: Optics and lasers in engineering. Volume 104(2018)
• Page Start: 204
• Page End: 219
• Publication Date: 2018-05
• Subjects: Defect inspection -- Anisotropic diffusion -- Radial basis function neural network -- Back-propagation neural network; Lasers in engineering -- Periodicals; Optical measurements -- Periodicals; Optics -- Periodicals; Lasers en ingénierie -- Périodiques; Mesures optiques -- Périodiques; Optique -- Périodiques; 621.36605
• DOI: 10.1016/j.optlaseng.2017.09.017
• Languages: English
• ISSNs: 0143-8166

Abstract

Highlights: High precision automated inspection and classification system for polarizing film. Median filter and anisotropic diffusion are used to remove the noises and sharpened the edge detail of defect region. The edge of defect region is completed by Canny edge detector and two-stage morphology processing. Maximum gray level, eccentricity, the contrast and homogeneity, gray level co-occurrence matrix are the inputs for RBFNN and BPNN classifier. Abstract: In order to improve the current manual inspection and classification process for polarizing film on production lines, this study proposes a high precision automated inspection and classification system for polarizing film, which is used for recognition and classification of four common defects: dent, foreign material, bright spot, and scratch. First, the median filter is used to remove the impulse noise in the defect image of polarizing film. The random noise in the background is smoothed by the improved anisotropic diffusion, while the edge detail of the defect region is sharpened. Next, the defect image is transformed by Fourier transform to the frequency domain, combined with a Butterworth high pass filter to sharpen the edge detail of the defect region, and brought back by inverse Fourier transform to the spatial domain to complete the image enhancement process. For image segmentation, the edge of the defect region is found by Canny edge detector, and then the complete defect region is obtained by two-stage morphology processing. For defect classification, the feature values, including maximum gray level, eccentricity, the contrast, and homogeneity of gray level co-occurrence matrix (GLCM) extracted from the images, are used as the input of the radial basis function neural network (RBFNN) and back-propagation neural network (BPNN) classifier, 96 defect images are then used as training samples, and 84 defect images are used as testing samples to validate the classification effect. The result shows that the classification accuracy by using RBFNN is 98.9%. Thus, our proposed system can be used by manufacturing companies for a higher yield rate and lower cost. The processing time of one single image is 2.57 seconds, thus meeting the practical application requirement of an industrial production line.