提出了一种基于神经网络学习特性和集合论的边缘分段算法(Merge Split),优点在于它的分段间断点、自由度满足自适应性,能在学习过程中实现而不必事先规定;进一步提出神经网络拟合不连续的图像边缘的方法,在无边缘数学函数形式的情况下,充分利用其学习函数的任意逼近和自学习性进行拟合,实践证明该方法可以克服以往边缘拟合方法的不足。 An edge segmentation algorithm based on learning characteristics and set theory of neural networks is proposed. The advantage of this algorithm lies in that its segmentation discontinuities satisfy the adaptability and can be realized in the learning process without prior specification. In this paper, a method of neural network fitting discontinuous image edge is proposed. With the form of edgeless mathematical function, this paper makes full use of arbitrary approximation and self-learning of its learning function to fit it. Practice has proved that this method can overcome the previous edge fitting method Insufficient.