具有旋转不变性的轮胎纹理特征提取

A rotation-invariant texture feature extraction method for tire pattern image

针对轮胎花纹图像,给出一种基于能量分布的曲波变换纹理特征提取算法。对轮胎花纹图像进行曲波变换,提取变换后子带的均值和方差作为特征值,构成特征向量,用以表示图像的纹理特征。计算各子带的能量,按大小排序,同时对特征向量进行循环位移,使能量最大的子带所对应的特征值位于特征向量首部,从而保证特征向量不因图像旋转而发生变化。对轮胎花纹数据库进行检索试验,结果表示所给方法的查准率为47.5%,优于小波变换算法的35.5%和曲波变换算法的41.17%。

A new image multi-scale texture feature extraction method with rotation invariance is proposed in order to improve the precision ot tire pattern image. Curelet transform is used to the tire pattern image to extract the mean and variance of each sub-band as the feature value. All feature values form a feature vector to represent the image texture features. The energy of each sub-band is calculated and sorted, and the feature vector with the largest energy feature value is recycling moved to the first place of the feature vector. Therefore the feature vector will not change with the image rotation. An experiment is carried out based on the tire pattern database.Experiment result shows that the precision of the proposed method is 47. 5%, which is better than that of 35.5% and 41.17% by the wavelet transform algorithm and the curvelet transform algorithm respectively.