Abstract: According to the characteristics of paper cut-out patterns exaggerative deformations，the seven moment invariants can be used as the eigenvector after the pretreatment of paper cut-out patterns image.The seven moment invariants normalized are input into BP neural network based on Levenberg-Marquardt algorithm and the trained network is used to be a classifier to realize the image pattern recognition.Experiments show that a relatively good recognition performance can be achieved to recognize paper cut-out patterns of exaggerative deformations.

摘要： 针对剪纸纹样艺术夸张变形的特点，将剪纸图像进行预处理，提取7个不变矩作为剪纸纹样的特征向量，采用LM算法优化BP神经网络，通过归一化后的不变矩对BP神经网络进行训练，应用训练后的神经网络作为分类器对剪纸纹样进行模式识别，实验证明该方法能够较好地识别有一定艺术变形的剪纸纹样。