一种基于多级空间视觉词典集体的图像分类方法  

An Image Classification Method Based on Multiple Level Spatial Visual Dictionary Ensemble

作　　者：罗会兰[1] 郭敏杰[1] 孔繁胜[2] 

机构地区：[1]江西理工大学信息工程学院,江西赣州341000 [2]浙江大学计算机科学技术学院,浙江杭州310027

出　　处：《电子学报》2015年第4期684-693,共10页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61105042;No.61462035);国家973重点基础研究发展计划(No.2010CB327900);江西省教育厅科技项目(No.GJJ13421)

摘　　要：针对单一特征时存在提取的信息量不足,对图像内容描述比较片面,提出将传统的SIFT特征与KDESG特征进行串行融合,生成一个联合向量作为新的特征向量.针对传统的视觉词典构造方法缺乏考虑视觉词汇在空间的分布特点,本文引入图像空间信息,提出了一种空间视觉词典的构造方法,先对图像进行空间金字塔划分,再把空间各子区域内的特征分别聚类,构建属于对应子空间区域的空间视觉词典.在图像表示阶段,图像各子区域内的特征基于其对应的空间视觉词典进行LLC稀疏编码,根据各子区域对图像贡献程度的不同,把编码后各子区域的特征向量赋予不同的权重加权处理,再连接形成最终的图像描述.最后,利用线性SVM进行图像分类,实验结果表明了本文方法的有效性和鲁棒性.Using single feature of images to describe the image content is one-sided because of insufficient information. To solve this problem, combing SIFF and KDES-G features to describe images is proposed by generating a joint vector as a new feature vector. Considering images＇ spatial information, an image classification method based on the spatial visual dictionary is proposed. Images are first divided into sub-regions according to spatial pyramid,and the spatial visual dictionaries are respectively constructed by grouping features of each region into a number of clusters. The features of each region are coded by LLC based on its corresponding dictionary, and then the coded feature vectors are given different weights according to the different contribution of each region. After that, the feature vectors of different regions are concatenated and regarded as the final image description. Finally, a linear SVM is used to classify images. Experimental results show that the proposed method has better performance and robustness com- pared with some state-of-the-art works.

关 键 词：图像分类 特征融合 空间视觉词典 LLC编码 加权处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]