The division-of-focal-plane polarization imaging system has a compact structure, small size, and high real-time performance, and simultaneously can achieve light intensity response of multiple polarization directions in single imaging. It is one of the research hotspots of polarization imaging. The planar structure reduces the spatial resolution of the image. To reconstruct a polarized image at full resolution and reduce the influence of the instantaneous field of view error, interpolation of polarized images is essential. To protect the tensor structure of polarized images, the algorithm for interpolation of polarized images based on non-negative sparse tensor factorization is proposed. First, according to the non-negative sparse coding theory, the four-channel polarization image block is tensor-decomposed. Second, the sparse representation is solved using nonlocal self-similarity constraints. Finally, the reconstructed image blocks are inverted mapped according to the sampling matrix to obtain a full resolution polarized image. Experimental results show that quantitative indicators and image reconstruction effects of the proposed algorithm are more accurate than that of current mainstream algorithms.