Abstract: An image stitching method for two images with parallax taking at different viewpoints is proposed based on the planar similarity among matching pairs of feature points. This method makes use of the property that feature points belonging to the same plane often share the same transformation matrix. By defining the planar similarity measure between feature point pairs, the missed matching rate is effectively decreased, and the image registration accuracy is improved. Subsequently, we adopt a graph cut-based algorithm to obtain the optimal stitching line. Experimental results on several datasets have demonstrated that our method outperforms the current state-of-art methods with a better anti-parallax robustness.