Discretizing tomographic operators is a crucial step in the design of reconstruction algorithms. In particular, for iterative methods, the projector and backprojector are usually assumed to be matched to guarantee convergence. In practice, the reconstruction task is challenging, especially in cone-beam geometry. To this end, in the context of CBCT with a flat panel detector, we propose to rely on an interpolating kernel based on the convolution of B-splines taking into account the sampling of the detector and of the volume. Matched CBCT projector/backprojector based on resampling transforms are designed.