基于Adaboost多特征融合的织物扫描图案识别

Fabric Scanning Pattern Recognition Based on Adaboost Multi-Feature Fusion

针对织物扫描图像中纱线纹理等的存在造成难以提取有效图案特征的问题,提出了一种基于多特征融合的图案识别方法。首先通过纹理抑制平滑滤波算法滤除织物扫描图像的纱线纹理,并进行灰度化;然后分别提取灰度图像的边缘方向直方图、最大稳定极值区域的SURF特征和灰度共生矩阵特征,建立样本图像特征库;最后以样本图像特征库特征为训练对象,通过Adaboost算法融合3类特征建立分类器,实现图案识别。实验结果表明,基于Adaboost的多特征融合织物扫描图案识别算法比单特征识别算法有较高的准确率。

It is difficult to extract effective pattern features due to the existence of yarn texture in fabric scanning images.For this problem,this paper proposes a pattern recognition method based on multi-feature fusion.Firstly,texture suppression smoothing filtering algorithm was applied to filter yarn texture of fabric scanning images and graying was conducted.Then,edge direction histogram of gray-level images,SURF features of maximally stable extremum regions and features of gray-level co-occurrence matrixes were extracted respectively,and feature database of sample images was established.Finally,by taking feature database of sample images as the training object,three types of pattern features were fused through Adaboost algorithm to establish the classifier and achieve pattern recognition.The experimental results show that the Adaboost-based multi-feature fusion fabric scanning pattern recognition algorithm achieves higher accuracy than singlefeature recognition method.