Time-division framework is commonly used in Mueller matrix polarimeters (MPs), which takes extra numbers of images at the same position in an acquisition sequence. In this Letter, we utilize measurement redundancy to raise a unique loss function which can reflect and evaluate the degree of mis-registration of Mueller matrix (MM) polarimetric images. Further, we demonstrate that the constant-step rotating MPs have a self-registration loss function free from systematic errors. Based on this property, we propose a self-registration framework, which is able to apply efficient sub-pixel registration skipping the calibration procedure of MPs. It is demonstrated that the self-registration framework performs well for tissue MM images. By combining with other powerful vectorized super-resolution methods, the framework proposed in this Letter has the potential to handle more complicated registration problems.