Improvement and application of a new FOA for image processing technology

This study proposes a novel image dual resolution optimisation algorithm (IDROA) that integrates a global immune-inspired search strategy with dual dictionary learning to meet high-definition image reconstruction and compression requirements. By combining an improved ResNet50 with a binary masking mechanism, the method overcomes the local optima limitations of traditional techniques and enhances reconstruction fidelity and compression performance. Experiments on Urban100, Set5, and Set14 show that IDROA achieves high reconstruction quality, with an average RMSE of 0.0401 and a compression ratio of 34.57%. Although its average processing time (2.605 s) and memory usage (523.32 MB) are slightly higher than conventional methods, it surpasses leading deep learning models such as SRGAN and ESRGAN in both accuracy and perceptual quality. The algorithm also demonstrates strong feature extraction capability, achieving an average similarity of 0.9153, highlighting its potential for high-quality imaging tasks under limited computing resources.