Image quality improvement for hybrid compressive light field display based on stochastic gradient descent with a mixed loss function

The multi-layer compressive light field (CLF) three-dimensional (3D) display suffers from image artifacts due to layer misalignment. In this paper, we propose to reduce the image artifacts by improving the intensity distribution uniformity of the layer images. To achieve this, a global optimization algorithm based on gradient descent (GD) is proposed. To further improve the image quality, a mixed loss function combining both subjective and objective losses is proposed for the optimization. By applying the subjective loss to enhance the perceptual quality, the 3D image details could be reconstructed better. The experimental results demonstrate that the proposed method significantly reduces artifacts and improves image quality compared to the conventional method.