A 28.8mW Accelerator IC for Dark Channel Prior Based Blind Image Deblurring
Published in IEEE Asian Solid-State Circuits Conference (A-SSCC), 2021
Recommended citation: P.-S. Chen, Y.-L. Chen, Y.-C. Lee, Z.-S. Fu, C.-H. Yang, "A 28.8mW Accelerator IC for Dark Channel Prior Based Blind Image Deblurring," IEEE Asian Solid-State Circuits Conference (A-SSCC), Nov. 2021.
Abstract: This work presents an energy-efficient accelerator that implements a dark channel prior based deblurring algorithm. With the proposed Laplace’s equation solver, the dark channel re-assigner and the conjugate gradient solver, a latency reduction of 76% for a 257x257 blurred image is achieved. The chip estimates a blur kernel for a full HD image in 1.7s and dissipates 28.8 mW at 65 MHz from a 0.65V supply. Compared to the state-of-the-art design, this work supports 4× and 2.9× larger image size and blur kernel size. It even achieves 4x [7.5x] higher normalized area [energy] efficiency, despite the better capability of image deblurring.