通过对可编辑逻辑门阵列(FPGA)内部乘法器及相应逻辑阵列资源的配置,将Gabor卷积的算法成功嵌入到邻域处理器当中,并通过高级语言的配置,使系统能够支持多种尺寸的卷积运算.实验结果表明,使用EP2C70F896C8型FPGA,在外部时钟为45MHz、内部时钟为180MHz的工作频率下,对于一幅120×90的灰度图像进行五尺度八方向的卷积运算(系数为16位复数)总耗时47ms,相对于使用Matlab软件做相同的卷积计算具有17.6倍的加速比.
This processor utilizes logic resources of FPGA to build up pipelined parallel convolution matrix,and use high level language to support any scale.Experiment indicates that Gabor filters with five scales and eight orientations(complex kernels) on 120×90 grey image take 47 ms on hardware platform with EP2C70F896C8 FPGA working at 45 MHz external clock,180 MHz internal clock,gaining a speedup ratio of 17.6 compared to Matlab.