匹配可信度是分析图象匹配质量的主要指标.针对归一化边缘强度互相关匹配算法,研究了基于神经网络的匹配可信度判别方法,即以参考图与若干实时图的匹配实验结果作为训练用样本数据,然后利用BP网络进行训练,再将训练后的网络用于匹配可信度的判别.通过实际卫片与航片图象对的匹配实验,证实了该方法的有效性.

Matching confidence is an important measure to analyses the quality of image matching. For normalized edge magnitude cross correlation matching algorithm, the measure of matching confidence based on neural network is studied. The training samples used to train a BP network are the matching results with the reference image and several sensed images. The trained network can be adopted to measure matching confidence. Experimetal results with real satellite images and aerial images prove the effectiveness of the method.