A Neural Network for Recognizing Rotation, Translation and Scale-Change Transformation of Patterns

在字符识别中，某些基本模式需要被正确识别为相同的模式，即使这些模式经历了(1)缩放变化、(2)平移和(3)旋转等转换。我们开发了三种类型的神经网络模型，能够学习和识别模式的尺度变化、平移和旋转。它们是(1)旋转不变学习模型；(2)模式旋转模型；(3)复制学习模型。这些都是三层神经网络。这三种模型的特点是网络结构简单，尺寸紧凑，大大减少了学习和识别所需的时间。

In character recognition, certain basic patterns need to be recognized correctly as the same patterns even if the patterns undergo transformations such as (1) scale-change, (2) translation, and (3) rotation. We developed three type of neural network model capable of learning and recognizing scale-change, translation and rotation of patterns. They are (1) A Rotation Invariant Learning Model ; (2) A Pattern Rotation Model ; (3) A Copy-Learning Model. These are all three layr neural networks. Each of the three models features a simple network structure of a compact size, and a substantially reduced time required for learning and recognition.