基于LBP-GLCM纹理特征提取的服装图像检索

Clothing Image Retrieval Based on LBP-GLCM Texture Feature Extraction

提出了一种针对服装图像检索的LBP-GLCM纹理特征提取方法,首先利用Uniform-旋转不变局部二值模式算法处理服装图像,得到不会因为旋转而改变的编码图像,然后获取此图像的灰度共生矩阵,采用能量、对比度、相关性和均匀性描述图像的纹理特征。实验数据表明,LBP-GLCM纹理特征提取方法具有良好的抗旋转性,且在采用欧氏距离作为相似性度量时,LBP-GLCM相较于GLCM纹理特征提取方法可以获得更高的检索效率和准确率。

A LBP-GLCM texture feature extraction method for clothing image retrieval is presented. First, the clothing image is processed by uniform invariant LBP to obtain the encoded image that does not change its features with the rotation. Then the statistics GLCM for this image is calculated and the image texture features with energy, contrast, correlation, and homogeneity are described. Experimental result indicates this LBP-GLCM texture feature extraction method possesses good anti-rotation ability. Moreover, LBP-GLCM texture feature extraction method retrieves clothing images more efficient and accurate than GLCM method when using Euclidean distance as the similarity measurement.