基于小波结构矩的图像识别算法

Image Recognition Based on Wavelet-Structure Moments

提出了一种基于小波结构矩的具有平移、旋转、缩放不变性的新型图像识别方法。小波结构矩是在小波矩的基础上通过改变图像函数的结构即几何矩的密度得到的。该算法结合了小波和结构矩的优点,不但实现了对图像特征的精细把握还增大了相似图像之间的距离。采用三次样条函数作为母小波,有效提取了图像的全局特征和局部特征。实验证明,小波结构矩比改进的Hu矩和结构矩具有更高的识别率,目前该算法已经成功运用到全自动金丝球焊机图像识别系统。

Wavelet-structure moments for image recognition were presented to deal with the images which were translated, scaled and rotated. This new theory was based on structure moments and wavelet transformation, after changing the structure of a image function, namely the density in geometric moments, the difference between images was enlarged, using mother wavelet of the cubic B-spine, both global feature and local feature can be extracted efficiently. Experiments demonstrate that wavelet-structure moments perform better than improved Hu moments and structure moments at recognition rate and correctness. And the method was applied into practical automatic golden ball wire bonder systems.