Adaptive Mask Optimization-Driven Hybrid Gan with Swarm-Based Tuning for High-Fidelity Image Inpainting: Results and Performance Analysis

Abstract

Image inpainting is a computer vision technique that aims to fill in missing or damaged areas of an image with plausible content. This paper presents the results and analysis of an Adaptive Mask Optimization-Driven Hybrid GAN designed to enhance image inpainting quality. The model integrates dynamic mask refinement with swarm-based hyperparameter tuning and a multi-component loss formulation to improve structural accuracy, texture realism, and computational efficiency. Evaluated on the Paris Street View dataset with cross-validation on three additional datasets, the proposed model achieved a PSNR of 34.72 dB, SSIM of 0.942, and EPI of 0.087, surpassing GAN-Inpaint, Painter Net, Hybrid Swarming Algorithm, and Swarm-Optimized Transformer GAN. Cross-dataset evaluation confirmed generalization across urban, artistic, and natural scenes. Ablation studies revealed that adaptive mask optimization and swarm-based tuning significantly improve perceptual quality, while computational analysis showed a balanced trade-off between accuracy and inference time. These results establish the model as a robust, efficient, and generalizable solution for applications in digital restoration, medical imaging, and remote sensing.