结合纹理和分布特征的遥感图像群目标识别方法

针对航空图像中的群目标提出结合纹理和分布特征的目标识别方法。该方法可以分为两步: 首先选取一组纹理特征,采用最大似然分类算法完成子目标区域的分割;然后基于分布特征从子目标区域中快速定位和识别群目标。实验表明,所选的纹理特征可以有效区分防护掩体与各种自然背景;提出的基于分布特征定位和识别目标的剪枝算法,与同类算法相比速度获得较大提高。对于多幅航空图像进行识别实验均得到满意的结果,表明这种方法可以有效的从复杂自然场景中快速识别出感兴趣的群目标。

A Teeming Targets Recognition Method Combining Texture and Distribution Features

Today's object recognition systems has become very good at recognition isolated objects. However, recognition in scenes with multiple objects is still problematic. Teeming objects consist of a number of similar sub-objects with definite distribution. In this paper, a promising recognition algorithm is presented for a kind of teeming objects in aerial images. The recognition is a three-stage process. First, a group of texture features are selected for figure ground separation. With these features, the raw aerial images are segmented using maximum likelihood classification method and the regions of interest are extracted. Finally, the distribution features are used to locate and recognize the teeming targets. We present a new pruning algorithm to locate the teeming targets based on its distribution features, which consuming less time than other algorithms. Experiment results review that the selected texture features can differentiate the sub-objects from nature background effectively. Recognizing experiments based on a set of aerial images indicate that the algorithm can recognize the teeming targets from complex natural scene effectively and efficiently.