基于自适应LBP和SVM的织物疵点检测算法

Fabric defect detection based on adaptive LBP and SVM

为准确提取不同种类织物纹理的特征,提出一种新的纹理特征描述方法——自适应局部二值模式(ALBP)。该方法为不同纹理结构创建相应的主要概率模式子集,避免了均匀局部二值模式(ULBP)使用同一模式集描述不同纹理而导致的描述不准确问题。在该算法基础上构建一种基于支持向量机(SVM)的织物疵点检测算法,将疵点检测问题转化为分类问题。实验结果证明,该算法不仅保持了传统局部二值模式(LBP)的旋转不变、多分辨率等特点,而且疵点检测结果在视觉上更加清晰、误检率更低、适用范围更广,SVM的优秀分类性能也有效地提高了疵点检测的准确率。

An advanced local binary patterns method was proposed to describe the main image features.Adaptive Local Binary Patterns （ALBP） method selected the frequently occurring patterns to construct the main pattern set,which avoids using the same pattern set to depict different texture structures in the traditional uniform local binary patterns.Based on the proposed method,an effective fabric defect detection algorithm of Support Vector Machine （SVM） was designed.First,the features of the training samples were extracted according to the set and were fed to SVM.Then the testing image was equally divided into detection windows from which ALBP features were also extracted and were classified by the trained SVM model.The experiments exhibit the detection effect of the proposed method is comparatively better than traditional LBP in terms of visual effect and detection accuracy.