Visual Words on Baggage X-Ray Images

X-ray inspection systems play a crucial role in security checkpoints, especially at the airports. Automatic analysis of X-ray images is desirable for reducing the workload of the screeners, increasing the inspection speed and for privacy concerns. X-ray images are quite different from visible spectrum images in terms of signal content, noise and clutter. This different type of data has not been sufficiently explored by computer vision researchers, due probably to the unavailability of such data. In this paper, we investigate the applicability of bag of visual words (BoW) methods to the classification and retrieval of X-ray images. We present the results of extensive experiments using different local feature detectors and descriptors. We conclude that although the straightforward application of BoW on X-ray images does not perform as well as it does on regular images, the performance can be significantly improved by utilizing the extra information available in X-ray images.