Polarization is one of the important physical properties of light. When targets on the Earth's surface or in the atmosphere reflect, scatter, transmit, or radiate electromagnetic waves, they generate specific polarization characteristics determined by their intrinsic properties. These polarization characteristics can be used to analyze target parameters such as shape, surface roughness, texture orientation, and physicochemical properties of materials. As a new-generation polarization detection technology, Division-of-Focal-Plane (DoFP) polarization imaging integrates polarizer arrays with focal plane detectors to achieve compact snapshot polarization imaging, demonstrating significant advantages in biomedical detection, environmental monitoring, and military reconnaissance. However, the DoFP detector structure causes reduced spatial resolution in imaging, and the varying intensity responses of adjacent pixels with different polarization orientations result in severe impacts on the reconstruction accuracy of polarization information due to Instantaneous Field-of-View (IFoV) errors. To achieve high-resolution imaging and reduce IFoV errors, DoFP polarization super-resolution imaging technology has become a research hotspot in this field. This paper systematically reviews DoFP polarization image demosaicking methods developed over the past decade, and analyzes development trends in this field, providing essential theoretical and technical support for advancing polarization imaging research. Starting from the fundamental theories of polarization imaging, this paper introduces typical polarization imaging systems: division-of- time, division-of- amplitude, division-of-aperture, DoFP, and metasurface-based ones. It then provides a detailed introduction and comparative analysis of the three major methodological frameworks for DoFP polarization super-resolution imaging, including traditional interpolation-based algorithms, mathematically-modeled optimization methods, and deep learning-driven intelligent processing techniques. Traditional interpolation-based algorithms essentially rely on mathematical modeling of neighborhood pixel correlations, reconstructing missing polarization information by exploiting spatial relationships between known pixels. Current technical frameworks focus on two core aspects: 1) constructing high-precision guide images to enhance edge preservation capabilities, and 2) designing adaptive weight functions to optimize noise robustness. Mathematically-modeled optimization methods achieve superior demosaicking results. However, as these methods fundamentally rely on iterative solutions to optimization problems, their iterative approximation mechanisms under non-convex optimization frameworks still face challenges in balancing computational complexity and convergence efficiency. Current research emphasizes two key priorities: 1) improving model efficiency through optimizations and parallel processing, and 2) developing decomposition models that incorporate polarization and spectral channel correlations while integrating polarization imaging processes to enhance modeling precision. Deep learning-driven intelligent processing techniques leverage neural networks’ powerful nonlinear representation capabilities to learn high-precision mapping relationships between mosaicked images and full-resolution images. On simulated data, their reconstruction performance far surpasses traditional and model-driven methods. However, due to the end-to-end black-box training paradigm of neural networks, the demosaicking optimization process lacks explicit physical interpretability. Challenges such as insufficient generalization in real-world scenarios and reliance on large-scale annotated datasets remain critical bottlenecks. Therefore, there is an urgent need to develop polarization image demosaicing networks with physical interpretability, strong generalization capabilities, and weakly-supervised (or unsupervised) learning properties. After over a decade of technological evolution, while significant breakthroughs have been achieved in DoFP polarization super-resolution imaging, future researches will focus on three major directions to meet high-precision imaging demands and advance technological frontiers: 1) Diffusion model-integrated polarization super-resolution imaging: By replacing "one-step" optimization with progressive refinement of super-resolution results, this approach aims to enhance the generalization capabilities of deep learning; 2) Deep unfolding model-based polarization super-resolution imaging: Combining explicit physical reconstruction models with the strong representation capabilities of deep neural networks, this framework constructs a fidelity-regularization alternating optimization architecture through deep unfolding, establishing an interpretability-driven and high-robustness polarization super-resolution system; 3) Building evaluation system for polarization image super-resolution: By developing a multi-modal quality assessment framework to establish a comprehensive evaluation benchmark tailored for polarization super-resolution imaging, which integrating three dimensions: polarization feature fidelity, scene adaptability, and algorithm robustness.