Edge-Aware Diffusion for Mobile Photo Enhancement: ASystematic Review and Comparative Latency Analysis

Although mobile devices capture and share the majority of images,image quality is often degraded due to low light, motion blur, limitedsensor capabilities, or underwater environments. Enhancing these imageson devices with limited resources is difficult because models needto deliver high visual quality while running in real time. While transformerand diffusion-based models can restore images effectively, traditionaldiffusion methods rely on multiple iterative steps making them slowand computationally expensive. Lightweight CNNs and hybrid architecturesrun faster, yet they sometimes miss fine details and struggle to maintainsharp edges. This review explores recent progress in edge-aware diffusionand lightweight image enhancement techniques for mobile devices followingthe PRISMA framework. We highlight methods that strike a balance betweencomputational efficiency and image quality, such as model pruning,one-step diffusion distillation, multi branch re-parameterized convolutionsand edge-aware loss functions. Research shows that diffusion modelscan now deliver near real time performance on smartphones while preservingfine structural details. At the same time, lightweight CNNs and Transformer-basedmodels offer fast, low-latency enhancement that make them ideal fordynamic scenarios like video processing. Hybrid and system-level strategiesincluding adaptive token reduction, patch-wise diffusion,and on-deviceGPU scaling further improve the feasibility of deploying these methodson mobile hardware.Our review shows that edge-aware strategies reliably enhance imagesin low-light, noisy or compressed conditions while hybrid approachesprovide a balanced trade-off between speed and quality. Complementingthis review, our empirical benchmarking reveals that while edge-awarediffusion offers superior detail, it incurs a 4x latency penalty comparedto lightweight CNNs on consumer hardware.