Concealed Weapon Detection in X-Ray Image Using Surf Features and Bag of Visual Words

Fadhil K. Al-Sheikh and Prof. Dr. Israa Hadi Ali

The process of detecting illegal objects in the X-rays of the luggage is a critical and important process, especially at the checkpoints, the entrances of the important buildings and also the airports. Therefore, a detection system has been developed to analyze X-ray images and detect threats to increase speed and accuracy, where some studies have shown that errors in manual detection range from between 80 and 90 %. The aim of the proposed system is to identify and discover important objects in the x-ray images of the baggage. The proposed system in this study has applied “Bag of Visual Words” approach which has been successfully used because it achieves high performance by reducing the number of extracted descriptors to maximum delegate’ ones. At first, speeded up robust features (SURF) has been applied on training image in order to extract descriptor for every key point, and then the visual vocabulary is constructed by K-means clustering algorithm. Thereafter, a histogram is created for the visual words that are a vector for the image feature. In the end, K-D Tree and Logistic regression were used as a classifier in order compared with each other to determine which classifier gave best results, as K-D Tree exceeded the Logistics regression by more than 10%. Each classifier has been trained using a feature vector for each image to distinguish images that contain illegal objects or not. The proposed system has been tested on the GD X-ray database containing multiclass: Razor, Knife, Handgun, Shuriken, and Built and recorded 97.1% accuracy as a position in the results. Bag of Visual Words achieved high performance for detection of the illicit object in X-ray security screening.

Volume 11 | 10-Special Issue

Pages: 708-719

DOI: 10.5373/JARDCS/V11SP10/20192861