地球信息科学学报 ›› 2019, Vol. 21 ›› Issue (4): 623-629.doi: 10.12082/dqxxkx.2019.180576

• 论文 • 上一篇    

Retinex图像增强在GPU平台上的实现

王浩1,2(), 王含宇1,2,*(), 杨名宇1,2, 许永森1,2   

  1. 1. 中国科学院航空光学成像与测量重点实验室,长春 130033
    2. 中国科学院长春光学精密机械与物理研究所,长春 130033
  • 收稿日期:2018-11-13 修回日期:2019-03-20 出版日期:2019-04-24 发布日期:2019-04-24
  • 通讯作者: 王含宇 E-mail:wanghao7600@163.com;hanyu112@126.com
  • 作者简介:

    作者简介:王 浩(1986-),男,吉林长春人,助理研究员,研究方向为航空数字图像处理。E-mail: wanghao7600@163.com

  • 基金资助:
    国家重点研发计划项目(2017YFB0503001、2016YFC0803000)

Implementation of Retinex Image Enhancement Algorithm on GPU Platform

Hao WANG1,2(), Hanyu WANG1,2,*(), Mingyu YANG1,2, Yongsen XU1,2   

  1. 1. Key Laboratory of Airborne Optical Imaging and Measurement, Chinese Academy of Sciences, Changchun 130033, China
    2. Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, China
  • Received:2018-11-13 Revised:2019-03-20 Online:2019-04-24 Published:2019-04-24
  • Contact: Hanyu WANG E-mail:wanghao7600@163.com;hanyu112@126.com
  • Supported by:
    The National Key Research and Development Program of China, No.2017YFB0503001, 2016YFC0803000

摘要:

伴随着无人机时代的到来,对海量数据处理的实时性要求越来越高。本文在GPU(Graphic Processing Unit)平台上实现了Retinex图像增强算法的并行处理,提升了Retinex图像增强算法处理高分辨率数字图像的处理速度。首先,通过数据合并访问和内存数据交互技术实现了数据的快速访问,缩短了数据在不同种类内存间的传输时间,提升了数据访问的效率;然后,采用内核指令优化和数据并行计算技术,实现了Retinex图像增强算法在GPU平台上的多核程序设计;最后,采用主机端和设备端的异步执行模式,在数据传输的同时进行内核数据的并行计算,通过任务级的并行进一步缩短了算法在GPU平台上的执行时间。研究表明,对于不同分辨率的图像,Retinex图像增强算法的处理速度相比于CPU平台均有数十倍的提高,如处理一帧分辨率为2048像元×2048像元的图像仅需要38.04 ms,算法的处理速度较CPU提高了40倍。

关键词: GPU, 图像增强, Retinex算法, 并行计算, 无人机

Abstract:

With the advent of the era of UAV,the real-time requriements for massive data processing are getting higher.Achieve parallel processing of Retinex image enhancement algorithm on the GPU (Graphic Processing Unit) platform, which improves the processing speed of Retinex image enhancement algorithm for processing high resolution digital images.Firstly, by data combine-accessing and memory data interaction technology realize fast access of data, shorten the transmission time of data between different kinds of memory, and improve the efficiency of data access. Then, using kernel instruction optimization and data parallel computing technology, the multi-core programming of Retinex image enhancement algorithm on GPU platform is realized.Finally, the asynchronous execution mode of the host and the device is used to perform parallel calculation of the kernel data while data transmission, and the execution time of the algorithm on the GPU platform is further shortened by the parallel of the task level. With the powerful parallel computing power of the GPU, the processing speed of the Retinex algorithm is greatly improved. For images of different resolutions, the processing speed of the Retinex image enhancement algorithm is tens of times higher than that of the CPU platform. Processing an image with a resolution of 2048×2048 pixels requires only 38.04 ms, and the processing speed of the algorithm is 40 times higher than that of the CPU.

Key words: GPU, image enhancement, Retinex algorithm, parallel computing, UAV