基于径向基神经网络的立体匹配算法

Stereo matching based on RBF neural networks

针对双目视觉中的图像立体匹配问题,提出了一种基于径向基神经网络的立体匹配算法。该算法提取图像的尺度不变特征变换(SIFT)特征建立特征匹配矩阵,对特征匹配向量进行约简,最后将约简的特征匹配向量输入径向基神经网络进行识别输出。仿真和实际图像实验表明,该算法的匹配正确率比标准的SIFT有所到提高。

Aiming at the stereo matching problem in computer vision, presented a novel stereo matching algorithm based on radial basis function neural networks. Firstly, extracted feature descriptors of the images using the SIFT（ scale invariant feature transform） descriptor. Secondly ,calculated the feature matching matrix between image pairs. Thirdly ,reduced the feature matching vector. Finally, passed the reduced feature vector to the RBF neural network to recognize whether the matching relationship was correct. Experiments on simulated and actual images show that the proposed algorithm outperforms SIFT in the aspect of matching precision.