一种用于小目标检测的可配置二维自适应预测器设计  

Architecture of a Configurable Two-Dimensional Adaptive Prediction Filter Used for Small Object Detection

作　　者：桑红石 沈绪榜[2] 陈朝阳[2] 傅勇[3] 

机构地区：[1]北京大学视觉与听觉信息处理国家重点实验室,北京100087 [2]华中科技大学图像识别与人工智能研究所图像信息处理与智能控制国家教育部重点实验室,湖南武汉430074 [3]华中科技大学控制科学与工程系,湖南武汉430074

出　　处：《小型微型计算机系统》2004年第8期1513-1516,共4页Journal of Chinese Computer Systems

基　　金：湖北省自然科学基金资助 ( 2 0 0 1ABB0 3 7)

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 。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.

关 键 词：小目标检测 自适应滤波 可配置性 图像处理 VLSI 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]