摘要
针对叶子图像的植物数据库的归类系统,提出了一种新的基于高斯混合模型特征函数的图像特征序列描述方法。定义了图像的高斯混合模型、特征函数及其性质,用自适应的方法把图像分解为K个模型,并在每个分量模型和混合模型上定义由频谱、相位角和功率谱组成的局部特征序列和全局特征序列。在中国科学院智能计算所的叶子图像数据集leaves(ICL)上进行了K-means归类实验,结果表明该图像描述方法比LBP局部综合特征和高斯混合密度函数有更好的归类结果。
For plant dataset categorization by leaf images,this paper proposed a new image signature method based on Gaussian mixture models(GMMs) and characteristic function.Firstly,it defined the GMMs for image,characteristic function and its property.Secondly,the image was divided into K models self-adaptively.Finally,it defined the local and global image signatures with spectrum,phase-angle and power spectrum of feature functions on each component model and mixture models respectively.And it developed a K-means categorization system to testify the effectiveness of the proposed method on the leaves(ICL) from intelligent computing laboratory of chinese academy of sciences.Experiment results show that the:proposed signatures can achieve better result than the classical LBP local combining features and GMMs density function.
出处
《计算机应用研究》
CSCD
北大核心
2012年第12期4740-4742,4746,共4页
Application Research of Computers
基金
江苏省高校自然科学基金资助项目(10KJB520004)
常熟理工学院校级项目(CITJGGN201117)
关键词
局部特征
全局特征
图像归类
混合模型
local feature
global feature
image categorization
mixture model