计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (19): 228-234.DOI: 10.3778/j.issn.1002-8331.2006-0304

• 图形图像处理 • 上一篇    下一篇

基于改进型LBP算法的植物叶片图像识别研究

李龙龙,何东健,王美丽   

  1. 1.陕西工业职业技术学院 信息工程学院,陕西 咸阳 712000
    2.西北农林科技大学 机械与电子工程学院,陕西 杨凌 712100
    3.西北农林科技大学 信息工程学院,陕西 杨凌 712100
  • 出版日期:2021-10-01 发布日期:2021-09-29

Study of Plant Leaf Image Recognition Based on Improved Local Binary Pattern Algorithm

LI Longlong, HE Dongjian, WANG Meili   

  1. 1.College of Information Engineering, Shaanxi Polytechnic Institute, Xianyang, Shaanxi 712000, China
    2.College of Mechanical & Electronic Engineering, Northwest A&F University, Yangling, Shaanxi 712100, China
    3.College of Information Engineering, Northwest A&F University, Yangling, Shaanxi 712100, China
  • Online:2021-10-01 Published:2021-09-29

摘要:

为了解决LBP算法抽取的纹理特征仅考虑了邻域像素的特征,忽略关键的局部和全局特征的问题,提出一种基于改进型LBP算法的WCM-LBP植物叶片图像特征提取方法。该算法融合了加权局部均值算法WRM-LBP和加权全局均值算法WOM-LBP,通过提取叶片基于区域的关键几何特征和纹理特征对LBP特征描述符进行加权改造,并采用加权局部均值和加权全局均值代替传统的中心像素点,最后将叶片图像的R、G和B通道颜色分量和灰度值作为特征输入矩阵进行图像分析。该算法结合特征加权的模糊半监督聚类算法(SFFD)应用于经典的Flavia、Swedish、Foliage以及自测图片集等4种植物叶片图像数据集中进行实验。实验结果表明,该算法具有很强的鲁棒性,能够有效区分机器视觉下植物叶片图像的关键性识别特征,有效解决叶片图像的分类识别中关键特征的描述问题。

关键词: 图像识别, 全局均值, 特征提取, 颜色分量, 植物叶片

Abstract:

In order to solve the problem of the classic Local Binary Pattern(LBP) algorithm that only the features of neighborhood pixels of the extracted texture features are considered and the key local and global features are ignored during the texture feature extraction process, an improved local binary pattern algorithm named WCM-LBP is proposed, which is combined with WCM-LBP and WOM-LBP, to extract the features from plant leaf images. It extracts the regional key geometry features and texture features of the leaf images to modify the LBP image descriptor, and uses weighted local mean value and weighted global mean value of the pixels instead of traditional center pixel, and finally extracts the color components of R, G, B channel and the common gray value for image analysis. Combined with the algorithm Semi-supervised Fuzzy clustering with Feature Discrimination(SFFD), this algorithm is applied to the experimental data such as classic Flavia, Swedish, Foliage datasets and self-test photo collections. The experimental results show that the algorithm has strong robustness, can effectively find out the key identification features of plant leaf images under machine vision, and effectively solve the problem of describing the key features in the classification and recognition of leaf images.

Key words: image recognition, global mean value, feature extraction, color component, plant leaf