Joint Denoising and Classification for Edge Devices

Overview

We jointly combine image classification and image denoising to enhance human perception of noisy images captured by edge devices like low-light security cameras. Using Neural Architecture Search, we optimize for latency, classification accuracy, and denoising performance simultaneously.

Key Contributions

• Novel integration of denoising and classification for edge deployment
• NAS-based architecture search optimizing multiple objectives
• Outperforms manually designed alternatives in both tasks
• Practical applications in surveillance, medical imaging, and industrial inspection

Links

• arXiv
• PDF