曲波变换用于磨粒图像不变矩的提取

曲波变换(curvelet)具有各向异性和良好的曲线奇异性表达能力。为了克服Hu不变矩和不变小波矩在表达铁谱磨粒形貌信息方面的不足,将曲波变换引入磨粒特征提取过程,并与Hu不变矩结合,提出一种基于曲波变换的磨粒图像不变矩提取方法。首先利用曲波变换将图像进行分解与重构,得到不同尺度的子图像;然后提取各子图像的Hu不变矩,获得图像的曲波不变矩;最后将该方法应用于典型磨粒识别,总识别率达到83.33%。实验结果表明,与Hu不变矩和不变小波矩相比,磨粒图像的曲波不变矩能更好地表达磨粒的形貌特征。

Invariant moment extraction by curvelet transform for wear particle images

The curvelet transform has the characteristic of anisotropy and the ability of good curve singularity expression. To overcome the shortage of invariant wavelet moments and Hu's invariant moments,curvelet transform is introduced into the wear particle feature extraction process and combined with Hu's invariant moments. Thus,an image invariant moment extracting method utilizing curvelet transform is proposed. First the wear particle images are decomposed and reconstructed by a curvelet transform,and their sub images of different scales are obtained. Then,the curvelet invariant moments are achieved by extracting Hu's invariant moments. Finally,the proposed method is applied for typical wear particle recognition,and a total successful recognition rate of 83.33% is accomplished. The experimental results indicate that compared with Hu's invariant moments and invariant wavelet moments,the curvelet invariant moments can better express wear particle appearance characteristics.