基于Grouplet-RVM的金属断口图像识别方法研究

Study on the recognition method of metal fracture images based on Grouplet-RVM

Grouplet变换是一种基于图像几何流最佳稀疏表示的正交变换,可以最大限度地利用图像的几何特征。关联向量机具有很好的泛化能力,能对类别的归属给出一种概率度量。结合Grouplet变换和关联向量机的各自优点,提出了一种基于Grouplet-RVM识别方法,提出的方法以Grouplet平均能量、Grouplet调和熵和Grouplet峭度为特征量,RVM为识别器,并成功地应用到金属断口图像识别中。实验结果表明,提出的方法是有效的,Grouplet峭度比Grouplet平均能量、Grouplet调和熵对断口图像的纹理变化更敏感,特别适于金属断口的特征提取。与小波-RVM识别方法相比较,提出的方法克服了小波-RVM识别方法只能获取图像有限的方向信息,取得了更高的识别率。和GroupletSVM识别方法相比较,Grouplet-RVM识别方法和Grouplet-SVM识别方法有同样好的识别率,然而,Grouplet-RVM的识别速度明显优于Grouplet-SVM识别方法,特别是随着训练样本的增加,这种优势越明显。

Grouplet transform is a new orthogonal transformation based on image geometric flow optimal sparse representation and can fully take advantage of the image geometry structure. Relevence vector machine （RVM） has good generalization ability and can give a probability measure for the classifications. Combing the advantages of Grouplet transform and RVM, a recognition method based on Grouplet-RVM is proposed. In the proposed method, Grouplet average energy, Grouplet harmonic entropy and Grouplet kurtosis are used as the image features, and RVM is used as a classifier. The proposed method has been successfully applied to the recognition of metal fracture images. The experiment result shows that the proposed method is very effective. In the three feature vectors, Grouplet kurtosis is most sensitive to the texture change of metal fracture and especially suitable for the feature extraction of metal fracture. Compared with the wavelet-RVM recognition method, the proposed method overcomes the shortcoming that the wavelet-RVM recognition method can only obtain the information in finite directions, and can achieve satisfactory recognition rate. Compared with the Grouplet-SVM recognition method, the proposed method and Grouplet-SVM recognition method both have the same good recognition rate. However, in recognition speed, the Grouplet-RVM recognition method is obviously superior to the Grouplet-SVM recognition method ; especially as the training samples increase its superiority is more obvious.