Low-resolution Touch Screen Surface Defect Detection Based On Feature Classification

Touch technology is the mainstream technology of natural human computer interaction at present. It has been used in the most intelligent terminal equipment. The quality of touch screen is very important for touch screen manufacturers. Currently, many companies still adopt manual inspection to detect the touch screen surface defects. However, with the production of touch screen increasing, manual inspection is difficult to meet the actual production requirements. Therefore, it is urgent for touch-screen manufacturers to develop an automated touch-screen surface defect detection system. Based on computer vision, image processing and other related technologies, we propose several automatic low-resolution touch screen surface defect detection methods in this thesis. The main investigation achievements of this thesis are as follows:(1) The touch screen surface images have the problems of the low-contrast between normal area and defective area, the similarity of the noise and slight defects etc. Aiming at the problems, we propose a method for low-resolution touch screen surface defect detection based on feature classification, and adopt Gabor feature to describe the touch screen images. The Gabor feature is used for the construction of a feature pool. After optimization, a dictionary can be obtained from the feature pool. For the testing images, sparse representation classification is applied for low-resolution touch screen surface defects detection.(2) Feature extraction and classification algorithm are critical for touch screen surface defect detection algorithms. Based on the local Gabor feature and global GLCM feature, a method for touch screen surface defects detection based on SVM is proposed. Also, proper classification models of SVM for different features are built respectively. Meanwhile, the principle of the parameter optimization of Gabor filter and the GLCM feature is proposed. According to the above methods based on feature extraction and classification, this thesis realizes automated low-resolution touch screen surface defects detection.Experimental results show that the accuracy of the method of Gabor feature-based sparse representation classification for the touch screen defect detection is up to 96%, and the method of the touch screen surface defects detection based on SVM classification gains a precision of 93.23%, which meets the production requirements of touch screen surface defect detection.