Model-based probabilistic relaxation segmentation applied to threat detection in airport X-ray imagery

This paper presents a new model-based probabilistic relaxation technique for segmentation. Model-based probabilistic relaxation labeling (PRL) uses high-level model characteristics to segment out objects in an image. Airport security involves X-ray imaging to examine the contents of the carry-on baggage. Traditionally, security officers must inspect large amounts of data, and in most cases discover no threat objects. As a result, they become less attentive and easily distracted, allowing for the possibility that threat objects may be overlooked. This type of situation is well suited for a computer vision system, which never tires, operates at a consistent level and could establish a minimum expected level of threat detection. Any computer vision problem can be divided into three stages: segmentation, feature extraction and feature evaluation. This paper presents a new method of segmentation. The segmentation phase of computer vision can be viewed as an input stage in which candidate regions of the image are identified, and if they meet basic conditions are processed by the remaining stages of the vision system.