Abstract: The Multi-Scale Retinex (MSR) image enhancement algorithm can produce best performance in most cases, but the computation load is very huge especially for large image. In this paper, an efficient approach was proposed to accelerate MSR image enhancement speed on Graphic Processing Unit (GPU) via Compute Unified Device Architecture (CUDA), and time consuming modules, such as multi-scale Gaussian filtering, log-differencing and dynamic compressing, were implemented in GPU. The experimental results show that the proposed method can reduce the time consumption significantly. As the image size increases, it can get a more than 100x speedup.

Key words: image enhancement, Multi-Scale Retinex (MSR), Compute Unified Device Architecture (CUDA)

摘要： 多尺度Retinex图像增强是一种基于色彩恒定理论的图像增强算法，算法增强效果好，但随着图像分辨率的提高计算时间显著增加。分析并利用计算统一设备架构(CUDA)图形处理器(GPU)的并行处理特性，提出了一种基于CUDA的多尺度Retinex图像增强并行算法，将多尺度高斯滤波、对数空间差分和动态范围压缩等计算非常耗时的模块采用并行方式放在GPU中进行计算。实验结果表明所提算法能显著提高计算速度，随着图像分辨率的增加，最大加速比超过100倍。

关键词: 图像增强, 多尺度Retinex, 计算统一设备架构