Polarization imaging, which captures polarization information from object surfaces, is widely used in industrial inspection, target recognition, and military reconnaissance. However, division-of-focal-plane (DoFP) polarization imaging systems based on mosaic filters have inherent limitations. The situation is more complex for color polarization imaging. The mosaic filter forces each pixel to capture not only a specific polarization angle but also one single color channel. This results in each pixel being severely deficient in both complete color and polarization data. Consequently, the raw data acquired are low-resolution and incomplete mosaic images, which heavily rely on post-processing to recover high-quality color polarization information. The traditional sequential processing paradigm of demosaicing followed by super-resolution (SR) suffers from error accumulation, low efficiency, and limited reconstruction quality. To address these issues, this paper proposes an end-to-end Dual-Stage Information-Complementary Network for joint demosaicing and SR of color polarization images. The network directly reconstructs high-resolution color polarization images from raw mosaic data through shared feature extraction and task-driven decoders. The core innovations include a Polarization-Aware Gated Dconv Feed-Forward Network, designed to achieve deep cross-modal fusion of polarization and intensity features, as well as a Task Interaction Module and a Multi-Scale Cross-Task Feature Reuse mechanism that effectively exploit the intrinsic complementarity between the demosaicing and SR tasks. Experiments demonstrate that our method significantly outperforms existing mainstream approaches across multiple metrics, while also enhancing the visual quality and polarization fidelity of the reconstructed images.