<p>Occlusion plays a critical role in accurate image analysis and feature identification in face recognition. Occlusions caused by random objects on an image can make it challenging to match the occluded image with the registered image in a database. To the best of the authors' knowledge, all well-known methods have focused on the problem of single occlusion in biometrics. To address the issue of multi-level occlusion, a novel feature-oriented GAN-based multi-occlusion removal framework (GMORF) is proposed. It consists of four components: face alignment using a spatial transformer network, a binary map generation with QUnet++, multi-occlusion removal through pixel-level similarity, and feature-level similarity enforcement using ResNet for improved inpainting. All components are integrated into an end-to-end network, and extensive experiments demonstrate GMORF's effectiveness in biometric verification with occluded faces.</p>