基于小波差分统计特征的纹理缺陷检测方法

纹理表面缺陷检测是机器视觉中一个重要研究课题,可广泛应用于纺织品、木材、陶瓷等产品的表面质量控制。针对统计纹理图像,提出了一种小波差分统计纹理特征缺陷检测方法。该方法首先应用小波基函数对纹理图像进行一级分解,将图像划分为互不重叠的子窗口,然后计算其差分统计纹理特征值,最后采用Mahalanobis距离分类器进行缺陷判决。实验不仅证明了该方法的有效性,并表明该方法速度快,可用于实时在线检测。

Texture defect inspection method using difference statistics feature in wavelet domain

Defect inspection for texture surface is one of important research problems in machine vision, and can be broadly used in surface quality control of different products, such as textile, lumber, ceramic tile and so on. A defect inspection method for statistics texture image, which uses texture difference statistics feature in wavelet domain, is presented. First, the texture image is decomposed by using wavelet base function, then the image can be partitioned into non-overlapping sub-windows and the texture difference statistic feature is computed. Finally each sub-window is classfied into defective or non-defective classes by using a Mahalanobis distance classifier. Experiments demonstrate the validity of our method, and show the potential possibility of real-time processing in an on-line inspection.