Division-of-focal-plane modulation is a powerful technique for real-time polarization imaging. This technique, however, suffers from the non-uniformity of the performance of linear polarization filters and photodetectors. We study the Stokes parameters reconstruction from the division-of-focal-plane modulation in the presence of the non-uniformity. Two reconstruction methods, named as least-squares and smoothing regularization methods, are proposed. The performance of the proposed methods are evaluated through Fourier analysis, numerical simulations, and experiments. The results indicate that the proposed methods can effectively mitigate the reconstruction errors and artifacts caused by the non-uniformity, and therefore may further facilitate the practical application of the division-of-focal-plane technique.