基于椭圆-双曲线马氏度量的图像分类算法  

Image classification algorithm based on elliptic hyperbolic mahalanobis metric

作　　者：鲍文霞[1,2] 阎少梅 梁栋 胡根生[1] 

机构地区：[1]安徽大学计算机智能与信号处理教育部重点实验室,安徽合肥230039 [2]偏振光成像探测技术安徽省重点实验室,安徽合肥230031

出　　处：《系统工程与电子技术》2018年第2期456-462,共7页Systems Engineering and Electronics

基　　金：国家自然科学基金(61401001;61501003;61672032);偏振光成像探测技术安徽省重点实验室开放基金项目(2016-KFJJ-001)资助课题

摘　　要：为了进一步拓宽度量学习在图像分类中的适用范围,同时提高分类的性能,本文提出一种基于椭圆-双曲线马氏度量的图像分类算法。该算法首先将颜色特征和局部二值模式描述的纹理特征相结合来表示图像特征;然后引入对样本数据具有更好的适应性的椭圆-双曲线度量,根据数据统计特性定义椭圆-双曲线马氏度量,给出椭圆-双曲线马氏度量学习算法,从而获取最优的度量矩阵;最后利用椭圆-双曲线马氏度量矩阵将样本变换到新的特征空间,从而降低特征各维度间的相关性,同时计算图像特征间的距离从而完成分类。实验表明该算法提高了图像分类的有效性。To widen the application scope of metric learning in image classification and improve the perfor mance of classification, an image classification algorithm based on elliptic hyperbolic mahalanobis metric is pro- posed. Firstly, the algorithm combines the color feature and the texture feature described hy local binary pat- terns （LBPs） to represent the image feature. Then, elliptic hyperbolic metric which has better adaptability to the sample data is introduced and the elliptic hyperbolic mahalanobis metric is defined according to the statistical characteristics of the data, and the elliptic hyperbolic mahalanobis metric learning is presented to obtain the op- timal metric matrix. Finally, the samples are transformed into a new feature space by using the elliptic hyper- bolic mahalanobis metric matrix to reduce the correlation between each dimension of the feature and complete the classification by calculating the distance between the features of the images. Experiments show that the pro posed algorithm improves the effectiveness of image classification.

关 键 词：椭圆-双曲线度量 马氏度量 度量学习 图像分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]