基于组合特征提取与多级SVM的轮胎花纹识别

摘要

Based on the important role of tire tread pattern in road traffic and criminal department, a novel approach of tire tread pattern recognition based on composite feature extraction and hierarchical support vector machine is proposed. The features of tire tread pattern are extracted by nonsubsampled Contourlet transform method and grey level co-occurrence matrix method respectively. The features extracted from the two methods are composited as tire tread pattern features, and five effective fea-tures are selected from all the features as the final features. The classification and recognition of the tire tread pattern is completed by using hierarchical SVM classifier and the five extracted features. The features extracted by the new method have the higher degree of separation among clusters. In addition, the classifying quality of hierarchical SVM based on decision tree is feasible and effective, which is significant for the correct classification and recognition of tire tread pattern.%基于轮胎花纹分类识别在交通与刑事部门的重要作用，提出了一种新的基于组合特征提取与多级SVM的轮胎花纹识别方法。分别采用非下采样Contourlet变换和灰度共生矩阵方法提取轮胎花纹特征；组合两种方法所提取的特征作为图像特征，并从中提取5个有效特征作为最终识别特征；运用提取的5个特征和多级支持向量机分类器完成轮胎花纹的分类识别。新的特征提取方法所得轮胎花纹特征分离度高，用决策树SVM分类器预测分类效果理想，对轮胎花纹的正确分类识别有着重要意义。