论文部分内容阅读
提出了一种新的基于前向神经网络和Ｈｏｐｆｉｅｌｄ 反馈神经网络的边界检测法，它分别探测每个象素点是否为边界点，便于实现边界检测的并行运算。首先设计了两层前向神经网络来增强和编码被检测象素点邻域的信息，然后利用增强和编码后的邻域图象作为Ｈｏｐｆｉｅｌｄ 反馈神经网络的输入，Ｈｏｐｆｉｅｌｄ 神经网络收敛时得到图象边界点。这种新的神经网络边界检测法所需的计算量比传统的Ｈｏｐｆｉｅｌｄ 网络边界检测法少得多，并增强了网络的抗噪声能力。整个神经网络是非监督学习的，为神经网络的训练提供了方便。
A new boundary detection method based on the feedforward neural network and Hopfield feedback neural network is proposed, which detects whether each pixel is a boundary point or not, which facilitates the parallel computation of boundary detection. Firstly, two layers of feedforward neural networks are designed to enhance and encode the information of neighborhoods of detected pixels. Then, the enhanced and encoded neighborhood image is used as the input of Hopfield feedback neural network. When Hopfield neural network converges, the image is obtained Boundary point. The new neural network boundary detection method requires much less computation than the traditional Hopfield network boundary detection and enhances the network’s anti-noise ability. The whole neural network is unsupervised learning, which provides convenience for the training of neural network.