Gun model recognition using geometric features of contour image

Global boarder customs seize large numbers of illegally smuggled guns annually, including large kinds of lethal aerodynamic guns. To classify the guns' types and recognize the models is beneficial and essential for the investigation of smuggling cases. Therefore, an automatic gun model recognition system with high efficiency is very important. In this work, we investigate the possibility that identifying a gun's model by its contour image. The procedure mainly involves acquiring a contour image using back-illuminated imaging and classifying the contour region based on geometric features. Four geometric features including area, circumference, maximum distance and Hu moment in-variants are extracted from the contour region. A combination of the above features is adopted for the gun's model recognition based on the Normalized Manhattan Distance rule. 79 water bomb guns that of 20 different models are used in the verifying experiment and 950 images are taken to form a dataset. Experimental results indicate that the method is able to classify and recognize the gun's model with a remarkably high accuracy of larger than 99%, which suggests that the contour differences between different models of guns can be detected by image classification. We expect this work will promote further studies on automatic gun model recognition and find potential applications in smuggling arms investigations.