Detection of a Real Thai National ID Card in Video Based

Identity verification isam andatory process when applying for various online services in Thailand. It requires presentation of your national ID card as evidence. The verification process requires users to capture a photo or video of their card. However, there is a risk of users attempting to counterfeit the card during this process. To mitigate this risk, printed holograms are applied to the card to prevent counterfeiting. The hologram provides a visual means of checking for authenticity. If a hologram is detected on the card, it serves as an indicator of a genuine card, facilitating the decision-making process in identity verification. This measure i s implemented to enhance security and prevent fraudulent activities. This research aims to apply deep learning techniques, utilizing the MobileNet SSD v2 as the backbone model, for hologram recognition on Thai national ID cards. We have developed a system specifically f or classifying holograms using grayscale images. In the training phase, the transfer learning method is employed, transferring pre-trained weights from the RGB model to initialize the grayscale model. The classification accuracy on the validation data indicates that the system's mAP (Mean Average Precision) at IoU (Intersection over Union) thresholds of 0.5 to 0.95 for the RGB model outperforms the grayscale model by 1.38%. However, the results for real card detection with grayscale video show an accuracy of 88.33%. These results demonstrate the success of developing a deep learning recognition system for detecting a real Thai national ID card in grayscale with high accuracy.