Abstract: Dim small targets detection in infrared image is one of the key techniques of infrared search and track system.To obtain more perfect detection result，some features of small targets in space and frequency domain are fused to detect those targets.In general，the infrared image can be divided into three categories：background，edges and targets.In space domain，the gray-scale feature is employed.The direction discreteness in frequency domain are extracted as a novel feature of target.Moreover，aiming to improve detection accuracy，it shows how to jointly make decision from gray-scale characteristics and direction discreteness.The targets are separated from background by exploiting the evaluation of Renyi’s information entropy at multi-feature values.A weighted variance growing method is proposed to filter suspicious targets and achieve accurate position by exploiting continuity and consistency of moving target in the image sequences and the variety growth of target.The proposed algorithm is flexible and background-insensitive.A series of evaluations on real battle-plan infrared video also show that the proposed method can effectively detect the small and weak targets.

Key words: infrared image, small and weak targets, direction discreteness, fusion, detection

摘要： 红外弱小点目标的检测是红外搜索与跟踪的关键技术之一。融合小目标在空域和频域中的各种属性，将更有利于目标的检测。红外图像中主要分为背景、边缘以及目标三类信息，目标在空域中局部能量较大。将图像小波变换，获取图像的多方向性分解。研究发现目标在高频中具有方向不敏感性。为了更好地检测目标，计算各点的局部能量比以及方向离散值，将以上特征融合，得到图像的多特征统计值。采用Renyi信息熵分割达到检测目标的目的。利用序列图像中目标运动的连续性和轨迹的一致性以及目标的方差增长性，提出一种加权的方差增长方法过滤目标集，实现候选目标的准确定位。该算法有较好的自适应性，并且对背景变化敏感性较小。通过真实红外图像弱小目标的检测，检验了算法的有效性。

关键词: 红外图像, 弱小目标, 方向离散性, 融合, 检测