A reference architecture for plausible Threat Image Projection (TIP) within 3D X-ray computed tomography volumes

Area of Science:

• Computer Vision
• Medical Imaging
• Security Technology

Background:

• Threat Image Projection (TIP) enhances human operator performance in X-ray security screening by overlaying threat signatures onto baggage images.
• Current TIP methods face challenges in 3D Computed Tomography (CT) due to complex volume segmentation and insertion point determination.

Purpose of the Study:

• To develop a novel 3D TIP approach for aviation security using CT volumes.
• To enable realistic and plausible insertion of threat objects within 3D CT baggage data.

Main Methods:

• The approach involves dual segmentation of threat and baggage volumes.
• Particle swarm optimization is used for determining threat insertion points.
• Metal artifact generation is incorporated for enhanced realism.

Main Results:

• Experiments used real airport baggage CT data to generate TIP volumes.
• Human evaluation graded 90.25% of generated TIP volumes as good, with minor flaws in 7% and 2.75% rated as bad.
• A TIP quality score metric was developed for automated evaluation.

Conclusions:

• The proposed 3D TIP method generates realistic and plausible threat insertions in CT baggage images.
• Generated TIP volumes are visually indistinguishable from real CT scans.
• The TIP quality score metric aligns well with human expert evaluations.