基于BP神经网络的景象匹配算法

A Scene Matching Algorithm based on BP Neutral Network

摘要:

针对以可见光图像为基准、红外图像为实测的景象匹配问题，提出了一种基于BP神经网络的景象匹配方法。该方法首先抽取满足平移、旋转、尺度不变性的Krawtchouk不变矩作为红外目标的特征描述并构造特征向量，随后，基于主分量分析法消除特征向量各个分量间的相关性，去掉多个对目标匹配识别贡献不大的特征的影响，形成描述目标的有效特征向量。接下来，构造三层BP神经网络，以有效匹配特征作为输入，匹配位置作为输出，按照经验公式确定隐层节点个数，基于样本集对红外目标在可见光基准图像中的匹配过程进行训练，最终形成智能化景象匹配识别器。实验结果表明，与常用的景象匹配算法相比，提出方法不仅具有更高的匹配精度和速度，而且鲁棒性好，能抵抗实测图像的旋转几何畸变。

Abstract:

Aiming for the problem of scene matching which takes infrared image as the actual data and the visible image as the referenced data, a matching algorithm is presented based on Back Propagation(BP) neutral network. Firstly, the features of Krawtchouk invariant moment for infrared target are abstracted. Then, the method of Principal Component Analysis (PCA) is used to remove the redundancy of the feature vector in order to form an effective description for target. Subsequently, taking the effective features as input character vector, a scene matching recognizer based on BP neutral network which had three layers is constructed. The experimental results show that the method given by this paper, compared with some classic matching algorithms, can not only has the high matching accuracy and fast matching speed, but also has better robustness which can resist the geometric distortion.