基于改进局部二值模式算法的中草药图像鉴别  

Herbs image recognition based on improved local binary patterns

作　　者：陶佳[1] 王芳[1] 沈红岩[1] 高媛[1] TAO Jia;WANG Fang;SHEN Hongyan;GAO Yuan(College of Information Science and Technology,Hebei Agricultural University,baoding 071000,China)

机构地区：[1]河北农业大学信息科学与技术学院,河北保定071000

出　　处：《河北农业大学学报》2020年第2期124-130,共7页Journal of Hebei Agricultural University

基　　金：河北省高等学校科学技术研究项目(2019211).

摘　　要：以颜色特征和纹理特征相结合的方法实现中草药显微图像的识别。将RGB图像转换为HSV图像,以HSV直方图对目标图像颜色特征进行提取;在传统局部二值模式中引入支持旋转不变特征的算子,并以均匀模式来描述目标图像的纹理,从而有效降低特征的维数;针对传统局部二值模式未考虑邻域像素分布特点而影响了纹理特征提取效果的不足,为其加入邻域点灰度值间的关系,对所有的相邻像素进行二值化;以Chi平方统计法对直方图进行相似性度量;为克服传统K近邻法的多优解问题,引入最短距离模式对最优解进行提取;最后通过优化算法来验证其集体性能,过程中结合训练集与测试集,证实优化算法的识别效果更佳。A new recognition method combined with color features and texture features is presented for microscope image of herbs.HSV space is selected and the transformation between RGB space and HSV space is presented for color feature extraction.To effectively reduce the dimension of features, the rotation invariant operators are introduced to the traditional Local Binary Patterns.In view of limitations of traditional method that the neighbor pixels are not considered, the relationship of intensity between neighborhoods points is added into the algorithm.A Chi Square statistic is taken for the similarity measure.The Shortest Distance pattern is used in order to solve the multiple optimal solutions. The training set and testing set were constructed to verify the effect of this model.The experiment results have shown that this algorithm is efficient and superior.

关 键 词：局部二值模式 中草药图像识别 K近邻分类器 显微图像 

分 类 号：TN911.73[电子电信—通信与信息系统]