[en] In this thesis we propose, implement, and evaluate algorithms improving spatial resolution in reconstructed images and reducing data noise in positron emission tomography imaging. These algorithms have been developed for a high resolution tomograph (HRRT) and applied to brain imaging, but can be used for other tomographs or studies. We first developed an iterative reconstruction algorithm including a stationary and isotropic model of resolution in image space, experimentally measured. We evaluated the impact of such a model of resolution in Monte-Carlo simulations, physical phantom experiments and in two clinical studies by comparing our algorithm with a reference reconstruction algorithm. This study suggests that biases due to partial volume effects are reduced, in particular in the clinical studies. Better spatial and temporal correlations are also found at the voxel level. However, other methods should be developed to further reduce data noise. We then proposed a maximum a posteriori de-noising algorithm that can be used for dynamic data to de-noise temporally raw data (sino-grams) or reconstructed images. The a priori modeled the coefficients in a wavelet basis of all the signals without noise (in an image or sinogram). We compared this technique with a reference de-noising method on replicated simulations. This illustrates the potential benefits of our approach of sinogram de-noising. (author)