Neural network-based denoising and capacity enhancement techniques for optical multi-image encryption systems

Abstract The importance of optical multiplexing encryption technology in information security is growing rapidly. As a result, optical multi-image encryption systems have become a major research focus. However, higher information density introduces challenges such as inter-image crosstalk and channel noise. These issues affect system stability, reliability, capacity, and practical value. To address these problems, this study constructs three optical multi-image encryption systems. These systems combine random phase and random amplitude as keys with wavelength, position, and angle as multiplexing methods. The performance of these systems and the noise in decrypted images are analyzed in depth. Based on the analysis, a neural network-based technique for denoising and capacity enhancement is proposed. Unlike existing studies focusing on encryption-end optimization, we propose a neural network-based denoising and capacity enhancement framework featuring three key innovations: an optical-deep learning collaborative architecture incorporating a decryption-end denoising module, achieving 5 to 10 fold capacity improvement while preserving physical security; an adaptive threshold control module overcoming fixed-threshold limitations of conventional DnCNN through noise-type-aware parameter adaptation; a multi-modal compatible network enabling cross-system noise suppression for wavelength, position, angle multiplexing mechanisms. Experimental results show that the proposed neural network maintains image integrity and accuracy while significantly increasing the capacity of optical multi-image encryption systems compared to other methods.