非正交二值子空间是一种重要的图像模板表示方法,能够高效地进行模板匹配。针对其生成过程耗时多难以实时应用的问题,提出了一种基于统一计算设备架构(CUDA)的非正交二值子空间模板表示的快速并行生成算法。该算法将模板生成过程划分为3个步骤,并根据每个步骤数据处理的特点,进行并行任务分解,极大地缩短了生成时间。实验结果表明,相比基于CPU生成相同的模板表示,该算法的速度提升了60~120倍。在应用于模板匹配时,该并行的模板表示生成算法能够很好的改进匹配效率。

Non-orthogonal Binary Subspace (NBS) is an essential image representation method,whose major advantage is to support high-efficiency image matching. However,generating NBS is itself time-consuming which makes the method difficult to be applied in real-time applications. In this paper,a parallel template-generating algorithm based on Compute Unified Device Architecture (CUDA) is proposed. The generating procedures are divided into three phases in our approach. For each phase,parallel task distribution is used to fully utilize the capability of the GPU. Experimental results demonstrate that our algorithm achieves 60~120 times speed-up,compared to the same template-generating task on the CPU. Additionally,we apply our algorithm on the template-matching problem,and observe significant improvement in efficiency.