Hardware Acceleration of Underwater Image Color Restoration Algorithm

Images captured in an underwater environment suffer from color degradation due to light scattering and absorption. This is a major obstacle to the effective utilization of underwater images, and accurate color restoration techniques are required. To address this issue, a physics-based color restoration algorithm, SUCRe (Scene’s 3D structure for Underwater Color Restoration), has been proposed. However, the SUCRe algorithm requires excessive computational resources as well as long processing times.In this study, we aimed to solve this problem by implementing the SUCRe algorithm on an FPGA to improve processing speed. In particular, it was found that the parameter optimization process within the algorithm was a bottleneck and required a large amount of processing time. Therefore, we accelerated this process by implementing the optimization process to the FPGA.As a result, compared to the original Python implementation, the optimization process was approximately 8 times faster in the C language implementation and approximately 190 times faster in the FPGA implementation.Regarding color restoration accuracy, we confirmed that the restoration performance on an underwater image dataset (including a color chart) was comparable to or better than conventional methods. This result indicates that we successfully shortened the overall processing time without sacrificing restoration performance, and suggests that further acceleration of the entire algorithm is possible by optimizing other processes within the algorithm.In future work, we plan to focus on speeding up and optimizing depth information acquisition methods, which are closely related to the entire algorithm, to further enhance the algorithm’s speed and improve restoration accuracy. Ultimately, we aim to realize a system capable of performing real-time color restoration at the point of capture.