Mura Defect Detection on Compact Camera Module (CCM) Using Metric Learning

The Compact Camera Module (CCM) is a device used for various compact electronic devices such as notebooks, smartphones, etc. Various defects occur in the manufacturing process, such as scratches, stamps, and mura. Most notably, mura defect detection is the most challenging issue because of how normal it appears. With this, various methods based on deep learning have been developed to detect mura defects. However, previous research assumes that there is a substantial amount of training data. Therefore, classification accuracy decreases in an environment wherein it is difficult to obtain a sample of an actual defect. This study proposes a metric learning-based mura defect detection method with higher classification accuracy than the previous semantic segmentation method in an environment with little training data. In the training phase, we obtained the center of the normal metric vector of the input image through the metric embedding model and metric loss, while in the test stage, we detected the mura defect based on the center of the normal metric vector. The experimental results show that the proposed method has higher detection accuracy than the previous method in an environment with few training data.