We propose an optimized initial weight scheme in a quantum-inspired neural network for compressing computer-generated holograms (CGHs). An optimized initial weight generation strategy is applied to accelerate the compression process. The pixel blocks’ complexity distribution of CGH is analyzed, and the parallel quantum neural network structure is used to compress the image pixel blocks. A deep convolutional neural network with residual learning is also adopted for improving the quality of the reconstructed image. The experimental results have shown that the compression iterations are reduced by using the optimized initial weight, and the reconstructed image quality of the compressed CGH is improved using the parallel quantum-inspired neural network structure and the deep convolutional neural network with residual learning.
A method for computer-generated hologram (CGH) compression and transmission using a quantum back-propagation neural network (QBPNN) is proposed, with the Fresnel transform technique adopted for image reconstruction of the compressed and transmitted CGH. Experiments of simulation were conducted to compare the reconstructed images from CGHs processed using a QBPNN with those processed using a back-propagation neural network (BPNN) at the optimal learning coefficients. The experimental results show that the method using a QBPNN could produce reconstructed images with a better quality than those obtained using a BPNN despite the use of fewer learning iterations at the same compression ratio.
An optimization scheme based on a genetic algorithm (GA) is proposed for kinoform synthesis. Unlike conventional optimization schemes, the initial kinoforms here are obtained by Fourier transform of the original image with random phase masks. The phase masks are then optimized by GA in order to reduce the reconstruction noise caused by amplitude negligence and phase quantization. Compared with the conventional methods of the genetic algorithm, in which optimization is directly performed to the kinoforms, the scheme can significantly improve the convergence and reduce the computation cost.
We propose a new scheme of computer-generated hologram (CGH) watermarking to resist rotation and scaling. To embed the inverse log-polar mapping of a mark pattern's CGH into a cover image, the twin image of the mark pattern can be directly reconstructed by fast Fourier transformation from the log-polar mapping of the watermarked image after rotation and scaling, not requiring a registration step in the extracting procedure. In an experiment, the information position of the twin image is located in the high-frequency domain and the redundant information of the low-frequency component is properly eliminated, so the contrast of the twin image is appropriately enhanced and the basic information of the mark pattern is effectively preserved to be recognized. The experimental results show that the mark-pattern's information can be effectively reconstructed when the watermarked image is scaled by 0.5 to 2 or rotated by any angle, so this watermarking scheme is effectively verified by experiment.