基于Criminisi算法的非周期性印花织物缺陷检测

由于非周期印花织物颜色多变、花型复杂、印花缺陷纹理及背景纹理之间对比度低,导致织物缺陷不易被识别,因此给出一种Criminisi算法。首先,采用频率谐调(frequency-tuned,FT)显著性算法对印花织物缺陷图像进行显著性增强,突出缺陷的边缘轮廓;其次,利用大津阈值法对显著图进行二值化处理,并将得到的二值图作为图像修复时的Mask掩码图;最后,运用Criminisi算法对原印花织物缺陷图像进行修复,将修复的印花图像减去原印花缺陷图像,获得缺陷区域。实验结果表明,该方法能准确分割出非周期性印花织物的缺陷区域,如漏墨、污渍等,且检测准确率达97.39%。同时,实验过程中不涉及参数的调节。

Detection of non-periodic printed fabric defects based on Criminisi algorithm

Non-periodics fabrics often have such defects as variable colors,complex patterns,low contrast between printing defects textures and background textures,which are not easy to be identified,so a Criminisi repair algorithm is proposed.Firstly,the frequency-tuned algorithm(FT)was applied to enhance the image saliency of the defect image and highlight the edge contour of the defect.Secondly,the Otsu threshold method was used to binarize the saliency map,and then the obtained binary map was used as the image repair Mask.Thirdly,the Criminisi algorithm was used for inpainting the original printed fabric defect image,the subtract operation was employed to process the repaired printed image and the original printed defect image to obtain the defect area.The experimental results show that this method can accurately segment the defect areas of non-periodic printed fabrics,such as ink leakage,stains,and the detection accuracy was 97.39%.At the same time,the adjustment of parameters is not involved in the experimental process.