针对用前视红外(FLIR)目标法识别复杂地面时, 存在无直接可用基准图、背景干扰严重、目标与背景灰度差异小等问题, 提出了一种基于区域模糊阈值的目标识别方法。首先, 在建立多尺度空间的基础上, 设计多阈值算法, 生成显著图; 其次, 引入基于图像模糊率的区域模糊阈值方法, 改进Itti模型, 构建候选目标筛选模型; 最后, 对检测结果运用积分归一化积相关(Nprod)算法进行精匹配, 确定识别目标。实验结果表明, 与Hausdorff距离算法相比, 该识别算法匹配率提高了近20%, 花费时间缩短了3/4; 与积分Nprod算法相比, 提出的算法匹配率提高了近40%, 时间缩短了1/2。结果显示, 对于复杂背景的前视红外目标, 该方法具有匹配率高、速度快、精度高等优点。

A target recognition method based on local fuzzy thresholds is presented to solve the problems of serious background interference, the absence of reference map for complex terrain objects and the low contrast between the target and the background. Firstly, a multi-threshold algorithm is designed and a saliency map is produced based on the establishment of multi-scale space. Then, the Itti model is improved and a candidate target filtering model is constructed by using the local fuzzy threshold method based on an image fuzzy rate. Finally, the detected results are precisely matched using Integral Nprod to determine the right one. The experimental results indicate that the match rate of the algorithm has increased nearly by 20% and 40% and the time consumption by 75% and 50% as compared with those of Hausdorff distance algorithm and Integral Nprod algorithm, respectively.In conclusions, the new algorithm has the advantages of high match rate, high speed and high accuracy for FLIR targets in complicated backgrounds.