Hybrid pixel detectors, originally developed for tracking particles in high energy physics experiments, have recently been used in material sciences and macromolecular crystallography. Their capability to count single photons and to apply a threshold on the photon energy suggests that they could be optimal digital x-ray detectors in low energy beams such as for small animal computed tomography (CT). To investigate this issue, we have studied the imaging performance of photon counting hybrid pixel detectors based on the XPAD3-S chip. Two detectors are considered, connected either to a Si or to a CdTe sensor, the latter being of interest for its higher efficiency. Both a standard 'International Electrotechnical Commission' (IEC) mammography beam and a beam used for mouse CT results published in the literature are employed. The detector stability, linearity and noise are investigated as a function of the dose for several imaging exposures (similar to 0.1-400 mu Gy). The perfect linearity of both detectors is confirmed, but an increase in internal noise for counting statistics higher than similar to 5000 photons has been found, corresponding to exposures above similar to 110 mu Gy and similar to 50 mu Gy for the Si and CdTe sensors, respectively. The noise power spectrum (NPS), the modulation transfer function (MTF) and the detective quantum efficiency (DQE) are then measured for two energy threshold configurations ( 5 keV and 18 keV) and three doses (similar to 3, 30 and 300 mu Gy), in order to obtain a complete estimation of the detector performances. In general, the CdTe sensor shows a clear superiority with a maximal DQE( 0) of similar to 1, thanks to its high efficiency (similar to 100%). The DQE of the Si sensor is more dependent on the radiation quality, due to the energy dependence of its efficiency its maximum is similar to 0.4 with respect to the softer radiation. Finally, we compare the XPAD3 S DQE with published curves of other digital devices in a similar radiation condition. The XPAD3 S/CdTe detector appears to be the best with the highest DQE at low frequency, although some improvements are expected to reduce the increase of noise with the counts statistics and to guarantee a better stability of the detector response.