基于CUDA的并行多尺度Retinex视频增强算法

多尺度Retinex图像增强算法增强效果明显,被广泛应用于图像和视频的增强处理中,但复杂的计算量限制了其在实时性应用中的推广,对于高清及多路视频的处理更是如此,因此研究其高速并行算法具有重要意义。本文以通用型GPU为基础,提出了一种基于CUDA的多尺度Retinex实时视频增强并行算法。根据算法各模块的耦合性将计算复杂的高斯滤波、对数空间差分及动态范围压缩等模块采用CUDA并行处理的方式实现,并利用视频序列之间的相似性降低多尺度Retinex算法的参数更新频率,以节省大量的计算耗时。实验结果表明所提算法能显著提高计算速度。

A Parallelized Multi-Scale Retinex Video Enhancement Algorithm Based on CUDA

The MSR(Multi-Scale Retinex) image and video enhancement algorithm can produce the best performance in most cases and is very popular in the field of video enhancement.However,MSR can not be applied and extended widely in real-time processing because the computation load is very huge especially for high definition and multi-channel videos.Thus the study on parallelized high-speed algorithms is tremendously significant.A parallel approach based on general GPU(Graphic Processing Unit) via CUDA(Compute Unified Device Architecture) is proposed in this paper in order to accelerate the speed of multi-scale retinex video enhancement.By implementing the computation complexity modules such as multi-scale Gaussian filtering,logarithmic domain differentiating and dynamic range compressing on GPU,and reducing the parameters updating frequency by using the similarity between consecutive frames,the computation complexity is saved a lot.The experimental results show that the proposed method can improve the computation speed significantly.