Deep learning-based edge enhancement of interferenceless coded aperture correlation holography without a point spread hologram

In interferenceless coded aperture correlation holography (I-COACH), it is still difficult to obtain high-quality edge-enhanced results due to the suppression of the modulation characteristics of the coded phase mask by cross-correlation operation. Here, we propose a deep learning-based edge-enhanced reconstruction method for I-COACH. Unlike the conventional method, which requires recording the point spread hologram of the system. Our proposed method, under the U-net architecture, directly establishes the mapping relationship between the object hologram of I-COACH and the edge-enhanced object. Thereby conveniently achieving high-quality edge-enhanced reconstruction compared with a complex iterative algorithm and a nonlinear edge enhancement algorithm. Moreover, the holograms obtained by different phase masks with different initial randomness and different occlusion interference are tested. When 90% of the pixels of the hologram are occluded, the network can still distinguish the feature of the object and achieve an acceptable edge-enhanced result, indicating strong robustness ability of the proposed method. Furthermore, the edge-enhanced imaging result of the 3D object at different depths demonstrates the applicability of the proposed network. Importantly, the proposed method provides a promising strategy of edge enhancement in 3D incoherent imaging and expands its application in pattern recognition and edge detection.