Cone-beam computed tomography (CBCT) is a standard nondestructive imaging technique related to the acquisition of three-dimensional data. This methodology interests a wide range of applications. An industrial CBCT system comprises an X-ray source and a flat-panel detector. Radiographic images are acquired during a rotation of the object of interest. A reconstruction algorithm leads to a volumic representation of the object and a post-processing routine assesses its validity. Accurate quantitative reconstruction is needed to perform an efficient diagsnotic. However, it is challenged by the presence of different artefacts coming from the acquisition itself. This thesis aims at analyzing and correcting those artefacts in a context of industrial micro-tomography. After an introduction to the physical and algorithmic background of CBCT, the artefacts are presented. Our study adresses two major artefacts: beam hardening and scatter radiations coming from the object and the detector. The second chapter reports on the state of the art in secondary radiation correction. A simulation model of the CBCT imaging chain is developed in a Monte Carlo environment. This model is designed to be realistic in order to get an accurate insight on the processes contributing to the final image formation. The third chapter focuses on the built and validation of the simulation tool. Monte Carlo methods are exact but prohibitively slow. Consequently, acceleration and optimization techniques are used to speed-up the calculations without loss of accuracy. A layer model of the flat-panel detector gives some insight on its secondary radiation behavior. More specifically, we demonstrate that a 2D description of the detector would be sufficient to compute its contribution. Our projection tool fits well with the real system. Finally, the last chapter describes our iterative correction method. The noisy initial reconstruction is segmented into different materials and densities and fed to the simulation framework. Beam hardening and secondary radiations are corrected via the volume reconstructed from the difference between acquired and simulated projections. This correction method is shown to be effective on both mono-material and poly-material objects.
