X-ray Computed Tomography (CT) is a non-destructive testing tool increasingly used by manufacturers, with growing interest in in-line testing applications. However, it is still struggling to establish itself as a standard technique due to long acquisition times. In order to meet industrial imperatives, recent research aims at reducing this acquisition time by limiting the number of views while keeping a good CT image quality. The rise of iterative reconstruction methods has made it possible to partially solve the Sparse-View CT problem thanks to the injection of regularisation terms. Furthermore, these methods also allow more freedom in the acquisition trajectories. In this work, we propose a method for reducing the number of projections without impacting the reconstruction quality by acquiring only the most relevant views. Indeed, not all views provide the same amount of information. We, therefore, present a technique for selecting views when the geometry of the inspected object is roughly known a priori. Our method is based on the Q-Discrete Empirical Interpolation Method (QDEIM) and considers the attenuation of the rays as an additional constraint.

X-ray Computed Tomography is a powerful non-destructive testing tool increasingly used by manufacturers to ensure the conformity of the produced parts. Despite growing interest, it is struggling to establish itself in online testing applications due to the large number of X-ray projections required to ensure a good reconstructed image. To reduce this number of projections from a few hundred to a few dozens while still getting satisfying reconstruction quality, we propose to infer a so-called mask on the volume to be reconstructed. By constraining the back-projection of the acquired X-ray projections only on this mask, corresponding to the voxels of the volume containing matter, iterative reconstruction algorithms, already very efficient at a low number of views compared to the traditional FDK, can better reconstruct an object, and with fewer computational resources. However, this technique requires a preliminary step: registering the experimental data to the a priori mask data. This paper presents a 3D/2D registration method based on Iterative Inverse Perspective Matching that registers a 3D CAD model to experimental projections. Then, we will explain how to construct the mask and use it during the reconstruction.

Cet article évalue le potentiel du convolutional sparse coding (CSC) pour réduire les artefacts dans les images 3D par tomographie rayons X dans le cas où peu de projections sont disponibles. La méthode CSC proposée est testée sur des échantillons métalliques fabriqués par fabrication additive, qui présentent des défis uniques pour les applications de débruitage en raison de la structure fine du matériau. Les résultats indiquent que le CSC surpasse les dictionnaires traditionnels en termes de performance de débruitage et de vitesse de calcul, ce qui en fait une méthode prometteuse pour les applications d'imagerie tomographique à grande échelle.