In medical applications, non-rigid image registration is often a first step for further actions, including more sophisticated image processing and pre-operative planning. In this study, we compared the performance of two non-rigid image registration algorithms. The first one, implemented in ANTs open source project, and second, a variational registration algorithm were used for registering MR lung images, that were processed afterwards to create functional lung images. To conduct a quantitative assessment of registered lung images, we used well known metrics (including Dice and Jaccard coefficients). The results were compared for thirteen MR lung image sequences recorded for five volunteers. Using this assessment, and visual evaluation of functional lung images that were derived from the registered ones, we found that variational image registration outperforms the ANTs registration algorithm, selected recently as highest ranking algorithm for human brain MRI registration. We found also, that variational registration algorithm is sensitive to regularization parameters, which means that variational registration algorithm may outperform ANTs algorithm only when optimal regularization parameters are chosen.