基于显著性检测的坯布疵点图像自适应分割方法

Adaptive segmentation method of grey fabric defect image based on saliency detection

针对坯布图像疵点相对占比小、易湮没在面料背景中,且疵点形状不规则、边缘难以有效分割的问题,提出了一种基于显著性检测的坯布疵点图像自适应分割方法.该方法采用视觉显著性方法分析疵点在坯布图像上的光强分布、形状及颜色差异特性,设计多视觉显著性特征函数,对显著性区域进行超像素标注;提出计算自适应分割阈值的方法提取疵点边界细节,对坯布图像上的疵点区域进行分割定位,实现疵点的精确检测.结果表明:该方法能够获得内部均匀致密且边界清晰明确的疵点图像区域,6类疵点的整体检测精度达到98.26%,提高了坯布疵点图像分割效果.

To address the issue of small defects blending into the fabric background due to the irregular shapes and the difficulty of effective edge segmentation,an adaptive segmentation method of grey fabric defect image was proposed based on saliency detection.This approach utilized visual saliency techniques to analyze light intensity distribution,shape and color difference characteristics of defects on grey fabric images,and a multi-visual saliency feature function was designed to label salient areas in super-pixels.Then,a method of calculating the adaptive segmentation threshold was proposed to extract the defect boundary details,and thereby segmenting and localizing defect regions on grey fabric images were carried on.Finally,the accurate detection of defects was realized.Experimental results show that the proposed approach can obtain uniform and dense defect areas with clear boundaries,and the overall detection accuracy of 6 types of defects reaches 98.26%,which can effectively improve the results of grey fabric defect image segmentation.