基于Gabor特征的木材表面缺陷的分块检测

Gridding Detection of Wood Surface Defects Based on Gabor Features

提出了一种新的木材表面缺陷的描述和检测方法,首先将木材表面图像划分成互不重叠的矩形块,即将木材图像矩阵进行分块;然后对每一块图像进行多方向多尺度Gabor变换,统计各个矩形块图像在不同尺度和方向上Gabor系数的均值和方差,将这些均值和方差组成一个描述矩形块的特征向量;为实现木材表面缺陷类别的检测,最后将块特征向量归一化后输入LS-SVM分类器,利用特征向量的相似度来进行缺陷的定位和识别。结果表明,该方法避免了传统检测方法需要进行图像分割的复杂性和局限性,它通过一个或多个矩形块的组合来定位缺陷,检测准确率超过91%。

We proposed a new method for wood surface defects description and detection based on Gabor features. Firstly, the wood surface image is divided into non-overlapping rectangular blocks. Then, every block of the image is decomposed by convolving with multi-scale and multi-orientation Gabor filters. Through statistical techniques, including mean and variance of Gabor coefficients inside each block, the block feature vector can be obtained to describe every block. Finally, the extracted feature vectors are normalized and inputted into the LS-SVM classifier to locate and detect the defects. Our method can avoid the complexity and limitations of image segmentation and the detection accuracy is more than 91%.