Tuesday, May 2025

04:20 PM - 04:40 PM

Room: LL20D

Session: Artificial Intelligence for Emerging Technologies and Applications

BDLUT: Blind Image Denoising with Hardware-Optimized Look-Up Tables

Distinguished Student

• Boyu LI

  The University of Hong Kong

Description:

We propose BDLUT that novelly combines blind denoising with hardware-optimized lookup tables (LUTs) for resource-efficient edge computing. BDLUT achieves superior denoising, outperforming state-of-the-art LUT methods by up to 2.59 dB on benchmark datasets using only 66KB storage. Our FPGA implementation demonstrates significant hardware efficiency, reducing logic resource consumption by over 10x and storage requirements by 75% compared to DNN accelerators, while achieving 57% higher speed than bilateral filtering methods.