A real-time Threat Image Projection (TIP) model base on deep learning for X-ray baggage inspection

Highlights: • We propose a real-time TIP model based on deep learning, which can predict the size type of the entire bag through the front part of bag. • We propose a mapping method that can map the real-world sizes of bags in optical images to the sizes of bags in X-ray images. • Our model employs a novel fusion method that can project the X-ray images of threat objects into X-ray images of bags. Real-time TIP is that X-ray security machine is scanning bag while projecting X-ray image of threat object into X-ray image of bag. When TIP starts, the front part of bag is only scanned. Threat object is very likely to be projected outside of bag. Therefore, we propose a real-time TIP model. This model contains a CNN-based classifier that can predict the size type of the entire bag through the front part of bag. After predicting the size type of the bag, X-ray image of the same type of threat object is projected into X-ray image of the latter part of bag. Moreover, we propose a mapping method, which can map the real-world size of bags in optical images to the size of bag in X-ray images. In addition, our model uses a novel fusion method to project the X-ray image of threat object into the X-ray image of bag.