• A calibrated multi-shot acquisition strategy overcomes structural undersampling by increasing the density of physically measured pixels before interpolation. • A pixel-accurate displacement protocol is validated using a nanometric hexapod, establishing a reliable physical-to-image-plane translation correspondence. • Sub-pixel phase correlation confirms integer-pixel displacements with residual errors below ±0.01 pixel across all spectral bands. • Multi-shot spectral densification with exclusively measured pixels progressively improves edge continuity, fine-detail reconstruction, and perceptual quality. • Physically measured pixels systematically outperform algorithmically estimated ones, as confirmed by comparison with state-of-the-art SD and PPID methods. One-shot multispectral imaging using Multispectral Filter Array (MSFA) sensors produces mosaicked data in which each spectral band is significantly under-sampled, limiting the reconstruction of spatial details and edge continuity. A fundamental consequence of this undersampling is that the majority of pixel values in each spectral band must be estimated through interpolation rather than physically measured, introducing reconstruction uncertainties that no algorithm can fully eliminate. To overcome this structural limitation, we propose a calibrated multi-shot acquisition strategy that increases the density of physically measured samples in each spectral band prior to any interpolation step. The approach relies on capturing several mosaics of the same scene through controlled integer-pixel translations of the MSFA sensor, executed with sub-pixel mechanical precision by a nanometric hexapod. To enable this acquisition strategy, a pixel-accurate displacement protocol is designed, calibrated, and experimentally validated using an Alio nanometric hexapod coupled to a PENTA/CAVIAR MSFA camera. A dedicated calibration procedure establishes the correspondence between the physical motion of the hexapod and the effective displacement measured in the image domain. Sub-pixel registration results show that a controlled sensor displacement of 0.32 mm corresponds to an effective translation of approximately one pixel in the image plane, confirming the reliability of the proposed calibration framework. The calibrated acquisition setup enables the generation of multiple spatially shifted mosaics whose complementary sampling patterns densify the initially under-sampled spectral bands. Experimental results demonstrate a progressive improvement in spatial continuity and fine-detail preservation as the number of calibrated acquisitions increases, with comparison against state-of-the-art demosaicking methods confirming that increasing the number of measured pixels systematically outperforms interpolation-based reconstruction strategies.
Yao et al. (Thu,) studied this question.