融合LBP纹理和局部灰度特征的材料图像分割

Material Image Segmentation Combined LBP Texture and Local Gray Level Feature

为了提高材料图像的分割精度,提出了融合局部灰度特征和LBP纹理的谱聚类分割算法。针对LBP算子无法区分邻域灰度差值幅度,提出几种改进的T-LBP算子,以表示图像纹理变化程度。构造邻域向量差用以描述局部特征;利用灰度直方图选取样本点;融合T-LBP特征与像素灰度特征及局部特征构造相似性矩阵;利用谱聚类算法进行图像分割;采用线检测方法抑制具有方向性的纹理噪声。对陶瓷材料图像和合成图像的实验结果表明,算法分割精度高、抗噪性强,具有较高的正确分类率。提出的融合LBP特征和灰度特征的谱聚类分割算法弥补了现有材料图像分割算法的不足,提高了材料图像的分割精度,适用于区域繁多、纹理复杂的材料图像;与其他算法实验结果的对比验证了该算法的有效性。

To improve precision of material image segmentation, based on spectral clustering method, a set of new algorithms combined local gray level features with Local Binary Patterns （LBP） are proposed. Considering that the LBP operator cannot efficiently distinguish the difference of gray magnitude of pixels in the neighborhoods,several threshold-LBP （ T-LBP） operators are proposed to show the change of image pixels. The difference of neighborhood vector is constructed to describe the local features,selecting sample points by gray level histogram, establishing the similarity matrix by combination of T -LBP features, gray features of pixel and local features, conducting the image segmentation by spectral clustering algorithm, and constraining the texture noise with direction by liner detection. The experiment for ceramic material image and synthetic image shows that the algorithm has high segmentation precision, strong noise resistance, and well correct classification rate. The proposed algorithm breaks through the drawbacks and improves the accuracy of material image segmentation, which is appropriate for various areas and complex texture of material images. The comparison among the proposed algorithm and other algorithm demonstrates the effectiveness of the former.