Diffractive imaging by multiple point detectors with learning empowering

Diffractive imaging outperforms traditional lens-based imaging in terms of enabling visualization of samples without bulky optics. However, most of the systems based on it necessitate coherent illumination, precisely engineered diffractive elements, a high resolution camera, and meticulously crafted algorithms. According to Huygens' principle, the optical field at any point in a diffraction field is the result of the superposition of contributions from all points in the input field. This fundamental theory suggests that original optical information can be retrieved by detecting several points in the diffraction field. Motivated by this, we proposed the method entitled diffractive imaging by multiple point detectors with learning empowering. Through this method, an object could be faithfully retrieved by utilizing a limited quantity of point detectors in conjunction with a neural network design. This method greatly relaxes the requirements on optical elements and coherent light sources, thereby significantly improving the accessibility and versatility of diffractive imaging applications.