构造高斯差分尺度空间的并行结构
Parallel Architecture for DoG Scale-space Construction
-
摘要: 提出了一种并行平滑机制,并基于该机制实现了构造DoG尺度空间的并行结构.在不影响匹配性能的情况下,与传统的结构相比,延迟线数量降低了84%,进而使得存储容量降低了90%左右.在131MHz的工作频率下,对于320×240的图像来说,完成高斯差分尺度空间的构造仅需要0.59ms.Abstract: In order to reduce the memory requirement caused by line buffers,we proposed a parallel architecture based on a parallel smoothing scheme to construct DoG scale space.Compared with the conventional architecture,the number of line buffers is reduces by 84%,which leads to the memory requirement dropped by 90%,in the similar false positive rate.It only takes 0.59ms to construct DoG scale space for a 320×240image at 131MHz.