In functional Magnetic Resonance Imaging (fMRI), the Hemodynamic Response Function (HRF) represents the impulse response of the neurovascular system. Its identification is essential for a deeper understanding of the dynamics of cerebral activity. In [1, 2], we developed a voxelwise approach i.e., based on a single time-course. In this paper, we propose an extension to cope with region-based HRF estimation. We introduce a spatial homogeneous model that assumes the same HRF shape for a majority of voxels within a given region-of-interest (ROI). A Least Trimmed Squares estimator is employed to select those voxels. A Bayesian HRF estimation is then performed with the corresponding time courses. Our approach is tested on real fMRI data to illustrate the gain in robustness achieved with the region-based estimate.