Pixel-segmenting imaging physically exceeds pixel-level sampling constraints via intra-pixel quantum-efficiency measurement enabled by evolutionary algorithms

Abstract Digital image sensors capture real-world scenes, yet their sampling resolution is limited by pixel physical dimensions—a constraint conventional deep learning-based super-resolution methods rarely overcome. For practical high-resolution imaging, here we show a pixel-segmenting imaging method. Leveraging a custom optical system and evolutionary algorithms to achieve accurate intra-pixel quantum efficiency measurement (a physical property) of digital image sensors, this approach physically exceeds pixel-level sampling constraints by reducing the minimal unit to 1/16 pixel and recovering high-frequency information missing from original images. Experimental results demonstrate the effectiveness of this method. At a 25 line pairs per millimetre cut-off frequency, its modulation transfer function and structural similarity index measure values are 2.3-fold and 3.1-fold those of the original images. Without modifying the original device structure, this method improves the angular resolution and pixel count of digital image sensors.