使用离散小波矩和遗传算法的目标识别

摘要

提出了一种连续小波矩的离散化算法,用于目标识别中的特征提取.因为可提取的小波矩数量较大,为简化分类器的设计,进一步提出了利用遗传算法(GA)进行自动的特征优化选择.仿真实验表明了离散小波矩的旋转不变性.基于瑞士联邦理工学院计算机视觉实验室的Obj_DB物体数据库,进一步测试了识别性能.结果表明,方法对任意选择的12个物体达到了100％识别率,优于无特征选择的方法和Hu矩特征方法.%For object recognition task, an algorithm to discretize the continuous wavelet moments for feature extraction is proposed.Because the quantity of calculated wavelet moments is very large, a method to use Genetic Algorithm(GA) to select appropriate moment features automatically is described.lt can also simplify the design of classifier.The experiment results show the rotation invariant property of discrete wavelet moments.Based on the ObjDB database provided by Computer Vision Laboratory of ETH-Zurich, the proposed method is evaluated also.For the 12 objects selected arbitrarily, the correct rate of recognition of this method achieves 100% and outperforms feature subset without GA selection and Hu moment features.