Diffractive deep neural network-based depth-of-field expansion without image restoration

In lens-based display systems, the depth of field (DOF) limitation of the lens often leads to blur and distortion of the reconstructed image. To overcome this limitation, we propose a depth of field extension method based on diffractive deep neural network. This method uses diffractive deep neural network to replace traditional diffractive optical devices (DOE) to extend the depth of field, and uses the Adam algorithm to optimize the phase distribution of the network, so as to achieve depth-invariant and concentrated point spread function distribution in the entire DOF range. Compared with other methods, our proposed method does not reduce the imaging quality and does not require a post-image restoration algorithm, which has greater advantages in integration and time cost.