Depth super-resolution is a process aimed at generating dense depth maps from sparse depth measurements, addressing the challenge of recovering detailed depth information and semantic context. Traditional methods in this domain have focused solely on enhancing sparse depth data. However, they often need to improve in reconstructing depth details and integrating semantic information, particularly in the absence of multi-modal data. This paper presents a novel approach to depth map super-resolution, leveraging a MobileNetV2-enhanced architecture that integrates multi-scale image guidance to reconstruct high-resolution (HR) depth maps from low-resolution (LR) counterparts. Our methodology employs a fusion of feature maps, utilizing the feature maps extracted from MobileNetV2, facilitating improved depth completion. We evaluate the proposed framework against established state-of-the-art methods using the publicly available Middlebury dataset to ensure a comprehensive performance assessment. The experimental results demonstrate significant improvements in both qualitative and quantitative metrics, underscoring the efficacy of our approach in enhancing depth map resolution.