基于C++AMP的计算机图像视频处理
|
摘要
随着计算机并行计算技术的快速发展,需要处理的数据量越来越大,提升并行计算的能力成为亟待解决的一个问题。在计算机图像处理的过程中,每一个环节可能都需要大量的计算,提出的C++AMP并行运算可以很好地提高计算机的处理效率。例如在视频监控内容处理时,为了获取到更加清楚的图像信息,通常会将图像放大,然而当图像放大后经常会看到边缘信息不清晰,图像异常模糊。能够获取一个清晰的图像边缘,对于信息的获取就非常重要了。对于视频监控图像模糊难以辨别的情况,提出了基于C++AMP的改进Canny算子的图像边缘信息处理方法。首先介绍了C++AMP的概念,以及现行的Canny算子处理方法,针对现行的Canny算子进行优化,并且对C++AMP并行化实现改进的Canny算子进行研究。然后采用改进梯度幅值的方法对Canny算子进行优化,通过对串行处理和改进Canny算子的并行化处理对比,可以看出并行运算的优势,图像越大,优势越明显,图像越清晰,边缘更多、更加明显,运算的时间也更快。
With the rapid development of parallel computing technology,the amount of data to be processed is also increasing,so the increase of parallel computing becomes an urgent problem to be solved.In the processing of computer image,each link may need a lot of computing,and the C ++AMP parallel computing can greatly improve the processing efficiency.For example in the surveillance video content processing,in order to get more clear image information,the image will be magnified. However,it will get a edge information after image magnification,not clear and abnormal blur.Therefore,acquiring a clear image edge is much important for information acquisition.For the monitoring image is difficult to identify,we put forward an image edge information processing method of improved Canny operator based on C ++ AMP.First we introduce the concept of C + + AMP and the current processing method of Canny operator. The current Canny operator is optimized,and the Canny operator improved by C ++AMP parallelization is studied.Then the improved gradient amplitude method is used to optimize the Canny operator.Form the comparison between the serial processing and the parallelization of Canny operator,we can see the advantage of the latter.The advantages are more obvious when the images are bigger.The edge will be wider and more obvious and the operation will get faster when the images are clearer.
With the rapid development of parallel computing technology,the amount of data to be processed is also increasing,so the increase of parallel computing becomes an urgent problem to be solved.In the processing of computer image,each link may need a lot of computing,and the C ++AMP parallel computing can greatly improve the processing efficiency.For example in the surveillance video content processing,in order to get more clear image information,the image will be magnified. However,it will get a edge information after image magnification,not clear and abnormal blur.Therefore,acquiring a clear image edge is much important for information acquisition.For the monitoring image is difficult to identify,we put forward an image edge information processing method of improved Canny operator based on C ++ AMP.First we introduce the concept of C + + AMP and the current processing method of Canny operator. The current Canny operator is optimized,and the Canny operator improved by C ++AMP parallelization is studied.Then the improved gradient amplitude method is used to optimize the Canny operator.Form the comparison between the serial processing and the parallelization of Canny operator,we can see the advantage of the latter.The advantages are more obvious when the images are bigger.The edge will be wider and more obvious and the operation will get faster when the images are clearer.
引文
[1]胡琼,秦磊,黄庆明.基于视觉的人体动作识别综述[J].计算机学报,2013,36(12):2512-2524.
[2]董荣胜.《九校联盟(C9)计算机基础教学发展战略联合声明》呼唤教育的转型[J].中国大学教学,2010(10):14-15.
[3]陈冠诚.C++AMP异构并行编程解析[J].程序员,2012(4):104-106.
[4]肖汉.基于CPU+GPU的影像匹配高效能异构并行计算研究[D].武汉:武汉大学,2011.
[5]丁鹏,陈利学,龚捷,等.GPU通用计算研究[J].计算机与现代化,2010(1):12-15.
[6]陈宏希.基于边缘保持平滑滤波的Canny算子边缘检测[J].兰州交通大学学报,2006,25(1):86-90.
[7]唐志文.浅析数字图像处理技术的研究现状及其发展方向[J].硅谷,2010(5):30.
[8]吴学明,李灿平.边缘检测算法在不同分辨率图像中的性能研究[J].计算机测量与控制,2006,14(2):166-169.
[9]王兰,吴谨.一种改进的Canny边缘检测算法[J].微计算机信息,2010,26(2):198-199.
[10]LIANG L R,LOONEY C G.Competitive fuzzy edge detection[J].Applied Soft Computing,2003,3(2):123-137.
[11]PELLEGRINO F A,VANZELLA W,TORRE V.Edge detection revisited[J].IEEE Transactions on Systems,M an and Cybernetics,2004,34(3):1500-1518.
[12]王蓉,高立群,柴玉华,等.综合Canny法与小波变换的边缘检测方法[J].东北大学学报:自然科学版,2005,26(12):1131-1133.
[13]朱仲涛,张钹,张再兴.图像关于边缘提取算子的微分不变性[J].计算机学报,1999,22(9):903-910.
[14]姚平.CUDA平台上的CPU/GPU异步计算模式[D].合肥:中国科学技术大学,2010.
[15]SANDERS J.GPU高性能编程CUDA实战[M].聂雪军,译.北京:机械工业出版社,2011.
[16]COOK S.A developer’s guide to parallel computing with GPUs[M].[s.l.]:M organ Kaufmann,2012.
[17]宗露艳,吴陈.一种改进的Canny算子边缘检测算法[J].现代电子技术,2011,34(4):104-106.
[18]张焕龙,胡士强,杨国胜.基于外观模型学习的视频目标跟踪方法综述[J].计算机研究与发展,2015,52(1):177-190.
[19]GOMEZ-LUNA J,GONZALEZ-LINARES J M,BENAVIDES J I,et al.An optimized approach to histogram computation on GPU[J].M achine Vision and Applications,2013,24(5):899-908.
[20]DESTREMPES F,MIGNOTTE M.A statistical model for contours in image[J].IEEE Transactions on Pattern Analysis and M achine Intelligence,2004,26(5):626-638.
[2]董荣胜.《九校联盟(C9)计算机基础教学发展战略联合声明》呼唤教育的转型[J].中国大学教学,2010(10):14-15.
[3]陈冠诚.C++AMP异构并行编程解析[J].程序员,2012(4):104-106.
[4]肖汉.基于CPU+GPU的影像匹配高效能异构并行计算研究[D].武汉:武汉大学,2011.
[5]丁鹏,陈利学,龚捷,等.GPU通用计算研究[J].计算机与现代化,2010(1):12-15.
[6]陈宏希.基于边缘保持平滑滤波的Canny算子边缘检测[J].兰州交通大学学报,2006,25(1):86-90.
[7]唐志文.浅析数字图像处理技术的研究现状及其发展方向[J].硅谷,2010(5):30.
[8]吴学明,李灿平.边缘检测算法在不同分辨率图像中的性能研究[J].计算机测量与控制,2006,14(2):166-169.
[9]王兰,吴谨.一种改进的Canny边缘检测算法[J].微计算机信息,2010,26(2):198-199.
[10]LIANG L R,LOONEY C G.Competitive fuzzy edge detection[J].Applied Soft Computing,2003,3(2):123-137.
[11]PELLEGRINO F A,VANZELLA W,TORRE V.Edge detection revisited[J].IEEE Transactions on Systems,M an and Cybernetics,2004,34(3):1500-1518.
[12]王蓉,高立群,柴玉华,等.综合Canny法与小波变换的边缘检测方法[J].东北大学学报:自然科学版,2005,26(12):1131-1133.
[13]朱仲涛,张钹,张再兴.图像关于边缘提取算子的微分不变性[J].计算机学报,1999,22(9):903-910.
[14]姚平.CUDA平台上的CPU/GPU异步计算模式[D].合肥:中国科学技术大学,2010.
[15]SANDERS J.GPU高性能编程CUDA实战[M].聂雪军,译.北京:机械工业出版社,2011.
[16]COOK S.A developer’s guide to parallel computing with GPUs[M].[s.l.]:M organ Kaufmann,2012.
[17]宗露艳,吴陈.一种改进的Canny算子边缘检测算法[J].现代电子技术,2011,34(4):104-106.
[18]张焕龙,胡士强,杨国胜.基于外观模型学习的视频目标跟踪方法综述[J].计算机研究与发展,2015,52(1):177-190.
[19]GOMEZ-LUNA J,GONZALEZ-LINARES J M,BENAVIDES J I,et al.An optimized approach to histogram computation on GPU[J].M achine Vision and Applications,2013,24(5):899-908.
[20]DESTREMPES F,MIGNOTTE M.A statistical model for contours in image[J].IEEE Transactions on Pattern Analysis and M achine Intelligence,2004,26(5):626-638.