基于改进Hopfield神经网络的图像特征点匹配算法

Image feature point’s matching algorithm based on improved Hopfield neural network

针对图像特征点匹配算法的运行时间呈指数增长的问题,提出了一种新的匹配算法NHop。该算法通过加入新的网络输入输出函数、点对间差异的度量和启发式选择目标点的方式,对传统的Hopfield神经网络进行了改进。新算法不仅解决了传统Hopfield神经网络运行时间长、能量函数易陷入局部极小点的问题,而且也有效地实现了图像特征点的匹配。实验结果表明,与传统的Hopfield神经网络相比,NHop算法的匹配速度更快、准确率更高,对于图像特征点的匹配效果更好。

In allusion to the problem that the run-time of image feature point’s matching will increase as the number grows exponentially, a new algorithm （NHop） to realize the problem of matching is presented. This algorithm improves Hopfield neural network by the addition of the method that a new input-output function of network, the difference measurement between two points and selecting target points during heuristic rules. The new algorithm not only solves the problem that traditional Hopfield neural network runs for a long time and its energy function is easy to drop into the local minimum, but also realizes the problem of image feature point’s matching effectively. The experimental result shows that compared with the traditional Hopfield neural network, the new algorithm NHop has a higher matching accuracy, a better matching effect on image feature point, and runs faster than traditional neural network.