改进频率调谐显著算法在疵点图像分割中的应用

Segmentation of fabric defect images based on improved frequency-tuned salient algorithm

为提高织物疵点分割精度,提出了一种用于织物疵点图像分割的改进频率调谐显著(FT)算法。首先,利用织物疵点和背景区域透光率的不同,将光源和相机分别置于织物两侧来获取图像,提高疵点区域对比度;其次,应用非局部均值滤波器(NLM)替代FT算法中的高斯滤波器,增强对背景纹理的平滑和降噪能力;研究发现NLM滤波器中滤波参数对疵点分割精度影响较大,提出了基于平均最大类间方差的参数优化方法;然后,将改进FT算法应用于疵点图像预处理,进一步提高疵点对比度;最后,使用最大类间方差法对疵点显著图进行分割。对2种不同织物疵点图像的分割实验结果表明,使用改进FT算法对粗经、竹节、结头、断纬、油污和破洞等常见疵点图像进行预处理,可显著提高疵点分割精度。

In order to improve the precision of fabric defects segmentation,an improved frequency-tuned salient（FT） algorithm is proposed for the segmentation of fabric image.Firstly,the light source and camera were placed on both sides of the fabric to obtain the image,and the contrast ratio of defect area was strengthened by the difference of transmittance between normal area and defect area.Secondly,the non-local mean filter（NLM） was used instead of the Gauss filter in the FT algorithm to enhance the cap ability of texture smoothing and denoising; and it is found that the NLM filter parameter has great influence on the accuracy of image segmentation.A method of parameter optimization using the average of inter-class maximum variance was proposed.Then,the improved FT algorithm was applied to the prepocessing of images to strengthen the contrast ratio of fabric defect area.Finally,OTSU algorithm was used to segment salient image of fabric defect.The experiments of image segmentation were carried out for two different fabrics.The experimental result shows that the segmentation precision of fabric defects,including slab yarn,knot,broken warp,oil stain,hole and so on,can significantly increase with the improved FT algorithm.