Abstract:

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.

Key words: video enhancement;multiscale retinex;compute unified device architecture(CUDA)