基于显著纹理特征的织物疵点检测方法

Fabric defects detection method based on texture saliency features

针对显著纹理背景下织物图像灰度级有限、对比度不明显致使目标疵点自动检测难度较大的问题,提出了一种用于显著纹理背景的织物疵点检测方法。鉴于Tamura纹理模型具有分辨能力强、旋转不变性以及算法鲁棒性强的特点,提出了多尺度度量局部纹理粗糙度的改进算法,以增强纹理分辨能力;结合织物疵点图像视觉显著性分析,基于局部纹理最佳窗口,通过提取与融合粗糙度、对比度和方向生成视觉显著性特征图,以显著突出织物疵点区。经TILDA织物纹理图库数据的实验测试结果表明,与其他相关方法相比,此方法在有效抑制显著纹理背景的同时,检测的目标疵点具有较好的一致性和完整性。

Owing to its low contrast,the defects of fabric images for background texture saliency are not very salient,and they are difficult to detect automatically. Aimed at this problem,a method for fabric defects detection based on texture saliency features is proposed in the paper. Firstly, in view of robustness of Tamura texture features,good discrimination and rotation invariance to texture,we present an improved local texture coarseness algorithm（ ILTCA） based on multi-scale calculation in order to further enhance discrimination to local texture. Then on a fabric image, coarseness, contrast and direction are calculated respectively based on optimal scale of local textures in accordance with ILTCA and three characteristic sub-maps are obtained,a salient feature map is formed by normalization and weighted fusion for difference sub-maps. Finally,comparing with the existing methods of fabric defects detection based on visional saliency feature,the comparing experiment results on the TILDA texture databases show that the proposed method can effectively isolate fabric defects from salient background texture and fabric defects detected has good homogeneity and integrality.