X-Ray Imaging/Inspection

X-Ray Imaging/Inspection

Kritiscan is a family of multi-energy X-ray baggage scanners designed and developed by Vehant Technologies. It consists of multiple models with varying inspection capabilities, different tunnel sizes, and varying levels of X-ray penetration. The machine displays color images as determined by the atomic number of the materials scanned. This helps the operator to easily differentiate between the types of materials inside the baggage such as metals, powder, plastic, liquid, etc.  

The research and development aspects of Kritiscan at Vehant are two folds: 1) to design and develop the hardware of next-generation X-ray machines, and 2) to enhance the inspection capabilities of the scanning machines by adding new features. The next-generation machines in X-ray imaging include body scanners, cargo container scanners, and baggage CT while the software development has its fundamental in core image processing algorithms specific to X-ray images that include image super-resolution, material discrimination, object detection, pseudo-coloring, image segmentation, image de-noising, etc.

X-Ray object detection

The state-of-the-art deep learning-based object detectors have made it possible to detect the threat objects while scanning. Neural networks trained on pre-specified classes like guns, knives, tools, etc are able to detect the object from these classes with reasonably good accuracy but non-quantification of the threats makes this problem quite challenging. There could be some new classes of potential threat objects which are not available in the training data or some instances of already existing classes that are totally new to the model. This makes it an open-world object detection problem and various verticals of deep learning like incremental learning, few-shot learning, learning without labels, etc can be explored.

X-Ray image super-resolution

The spatial resolution of X-ray images is bound by the minimum physical size of the detectors. As a result, it sometimes becomes difficult to capture very thin entities, like wire, having a thickness less than the detector size and the generated images may also appear smeared. With the increasing demand for high-resolution images, it has become necessary to explore super-resolution algorithms to enhance the image details. State-of-the-art generative adversarial networks show promising results for image super-resolution task. These algorithms can be explored in a model compression framework to work in a real-time scenario.

Material discrimination in X-Ray images

Dual enery X-ray scanning systems have made it possible to determine the effective atomic number of the scanned materials and categorize it further into different categories: organic, inorganic, and metal. Moreover, the overlapping of a threat object with other materials potentially changes the object’s properties and makes the threat recognition task difficult.  In material discrimination, an object of interest is separated out from the background using various image processing algorithms like background subtraction, basis image decomposition, object segmentation etc.

CT Reconstruction

Computed tomography (CT) is a process of capturing  X-ray projection measurements through objects’s cross-sectional area at different angles for image synthesis. In the literature, the algorithms for image reconstruction from these measurements are mainly divided into two categories: analytical and iterative approaches. Moreover, baggage CT  has to face different challenges as compared to medical CT.   The spatial resolution of the generated image is very high in medical CT, but in baggage CT, the spatial resolution can be compromised to some extent to achieve higher scanning speed. Furthermore, no prior information of the scanned material is available. These issues make the reconstruction task challenging and open various research verticals in this domain.