T2 mapping provides valuable tissue-specific MR information. To enable shorter scan times, multi spin-echo (MSE) sequences are commonly used but the achieved T2 accuracy using conventional mono-exponential fitting is poor. Improvements are possible by matching the measured signal to a pre-computed dictionary. Although simultaneous B1+ estimation is feasible, previous work demonstrated a bimodal behaviour. We investigate further improvements in B1+ accuracy using an iterative pixel-neighborhood based method (the Fusion Bootstrap Moves Solver), comparing different levels of spatial regularization. Improved B1+ accuracy and recovery of spatially smooth maps was demonstrated both in simulated and in-vivo brain data.