归一化方法用于PSRI目标识别

本文介绍了一种解决位移，大小和旋转不变性的归一化方法，图像的归一化由矩来实现，其中解决方向不变性采用了二点法。为了提高归一化后的二值图像相互之间的正交性，文中采用了反归一化方法，在此基础上应用ＦＤＯ神经网络进行目标分类，实验结果证明，此系统具有较好的分类能力。

^Invariant Image Recognition by Normalization

In this paper a normalization method for invariant image recognition is proposed. Image normalization is completed by moment and normalization for rotation is finished by two-points which is invariant with respect to image. These two -points are obtained by mapping the image. In order to improve the orthogonality between images, an antinormalization in used. At the basis of the above processing FDO (Full Domain Optimum) neural network is suggested to classify image objects. Experiment results show the high performance of this invariant image recognition system.