基于轮廓波变换的织物疵点分割

Fabric defect segmentation based on contourlet transform

为解决织物疵点分割中边缘保持能力差的问题,提出了一种基于轮廓波(contourlet)变换的织物疵点分割算法。采用拉普拉斯金字塔(LP)与方向滤波器组(DFB)实现织物图像的分解,将分解得到的子带系数进行基于区域能量极大值算法的轮廓波重构,对重构后图像进行全局阈值分割及形态学运算从而得到清晰完整的疵点轮廓。实验中对双纬、断纬、断经、破洞和油污等5类常见疵点图像进行分割,分割结果表明,该方法可实现方向疵点和局部疵点的正确分割,能够适应织物疵点方向性变化,为织物疵点的自动检测提供了新思路。

To solve the problem of poor edge holding ability in fabric defect segmentation,a fabric defect detection scheme based on contourlet transform was proposed.First,the Laplacian pyramid （LP） and directional filter banks （DFB） were used to achieve the decomposition of fabric image.Then,the decomposed sub-band coefficients were reconstructed based on regional energy maxima.Finally,clear and complete defect profiles were obtained by thresholding segmentation and morphological processing.Experimental results including five kinds of common defects,such as double weft,broken weft,broken warp,holes and oil,showed that the proposed method could achieve correct segmentation on directional defects and regional defects and adapt to directional changes of fabric defects.It provided a new idea for the automatic detection of fabric defects.