【摘要】 分析了用于图像中小目标检测的自适应预测器的支撑区域与其检测性能之间的关系 ,认为支撑区域应按照被处理图像统计特性进行设置 ,以正确区分图像中的目标与背景成分 .在此基础上 ,提出了一种基于 TDNL MS(Two Di-m ensional Normalized L east Mean Square)算法的支撑区域可配置的自适应预测器结构 ,通过设置适当的支撑区域 ,该预测器不仅可用于处理具有不同统计特性的图像 ,而且可以在一定程度上保持对成像面积逐渐变化的小目标的检测能力 .本文提出的预测器结构 ,只需要在每个抽头内部增加简单的控制逻辑 ,就可以实现支撑区域的任意配置 ,是一种较为理想的设计方案更多还原

【Abstract】 In this paper, we discuss the relation between adaptive filters support region and the small object detection performance, and propose that the support should be configured according to the environment stochastic characteristic. A configurable two dimensional adaptive digital filter architecture based on two dimensional normalized least mean square (TDNLMS) algorithm is proposed in this paper. This architecture can be used as pre whitening filter to enhance the detectability of small objects in digital image. The proposed configurable architecture can be used to process image series with different stochastic characteristic, and keep detection performance of the small objects whose spatial spread are changing during detection process. To configure the support region of the adaptive filter, only some simple logic are needed in every tap.更多还原