基于多级分组排序网络的实时大窗口二维中值滤波器设计

Real-time large window-sized 2D median filter based on multi-phased grouping and sorting network

二维中值滤波器是先进视频跟踪器设计的一项重要内容,对提高红外成像跟踪器的预处理性能有重要意义,大窗口中值滤波器的设计是实时图像处理滤波器族中最为复杂的技术之一。针对窗口尺寸大于3×3的实时中值滤波器设计问题,介绍了一种基于FPGA实现大窗口实时二维中值滤波器的设计方法。通过三级分组排序网络逐渐缩小中值的分布范围,将问题化简为9点中值求解。设计了专用并行流水线结构,实现了5×5二维实时中值滤波器。实验结果表明,滤波器占用资源少,数据吞吐率高,时延小,可进行本位运算,综合指标较高。该方法可推广到诸如7×7,9×9等更大窗口的二维中值滤波器设计中。

2D median filter is an important part in the design of the advanced video tracker. In particular, it can improve the pre-processing performance of the infrared video tracker. However, the design of the real-time median filter with the large window size is one of the complex technologies in real-time image processing filter families. In order to deal with some difficult problems existing in the design of the large window-sized （i.e. move than 3×3） real-time median filter, a new method for the design of a real-time large window-sized 2D median filter was presented based on field programmable gate array （FPGA）. The multi-phased grouping and sorting network were adopted to reduce the median distribution range of the window, and then the median calculation of the large window was simplified into that of the 3×3 window. Special parallel pipeline structure was designed to fulfill a 5×5 2D real-time median filter. The experimental results show that the method has lower resource consumption, shorter time delay, higher throughput rate, and in-place operation. Therefore, the greater comprehensive index of the filter is achieved. In addition, it can be applied in the design of larger window-sized median filters such as 7×7 2D real-time median filter, 9×9 2D real-time median filter and so on.