Crosstalk suppression via mirrored patterns in parallel single-pixel imaging

Non-visible imaging is an essential technique in advanced instrumentation and measurement systems, yet the high cost of high-resolution non-visible focal plane arrays (FPAs) limits its application. Parallel single-pixel imaging (PSPI) offers a cost-effective solution by reconstructing high-resolution images from low-resolution images modulated with a sequence of patterns. However, optical aberrations, especially defocus, introduce sub-image crosstalk, resulting in unnatural artifacts that degrade reconstruction quality. Existing suppression methods, such as transfer function matrix, deep learning approach and pattern design strategy, are often constrained by frequent recalibration requirement, high computational cost and residual random noise. To overcome these limitations, a mirrored pattern strategy is proposed to inherently suppress crosstalk by reshaping the artifacts to resemble the defocused spots. By transforming the recovery of original pixel values from an ill-posed inverse problem into a well-posed problem, the proposed method enables a precise solution without complex calibration or extensive training. Simulations and experiments demonstrate that the proposed method significantly improves reconstructed image quality and provides a robust and efficient approach to enhance the spatial resolution of non-visible imaging systems, achieving an average 27.7% reduction in root mean square error (RMSE) and an average 9.1% increase in peak signal-to-noise ratio (PSNR).