基于多特征融合与稀疏分类的图像检索算法  

A New Image Retrieval Algorithm Based on Multi-feature Fusion and Sparse Classification

作　　者：张光辉[1] 徐学红[2] 

机构地区：[1]宿州学院数学与统计学院,安徽宿州234000 [2]河南牧业经济学院信息与电子工程学院,郑州450044

出　　处：《控制工程》2016年第11期1796-1801,共6页Control Engineering of China

基　　金：安徽省高校优秀青年人才基金(2012SQRL201)

摘　　要：目前基于内容的图像检索研究中,图像特征提取的方法和相似性度量是影响检索准确性的关键因素。为破解当前基于内容的图像检索的语义鸿沟难题,提出基于多特征融合与稀疏理论的图像检索算法R_GLCM_SRC(Gray Level Co-occurrence Matrix,Sparse Representation-based Classifier)。R_GLCM_SRC利用颜色矩、灰度共生矩阵分别提取了图像的颜色和纹理特征作为综合特征,通过稀疏分类方法将底层特征分类问题转换为稀疏系数向量分类问题,从而实现图像分类检索。实验结果表明,R_GLCM_SRC提取的图像综合特征能够很好的实现图像的检索,在稀疏空间可有效缓解语义鸿沟问题,相对于最近邻分类检索算法,R_GLCM_SRC采用的稀疏分类算法检索效果更为稳定。At present, image feature extraction and similarity measure are the key factors that affect the retrieval accuracy in the research of content-based image retrieval. In order to solve the problem of the semantic gap in current content-based image retrieval systems, a new image retrieval algorithm based on multi-feature fusion and sparse classification called RGLCMSRC is presented in this paper. RGLCMSRC uses color moments and gray level co-occurrence matrix to extract color and texture features of the image as a comprehensive feature, and the classification of the underlying feaatres can be turned into a sparse coefficient vector classification problem by means of the sparse representation classification method, then we can realize image classification and retrieval. The experimental results show that the image comprehensive features extracted in RGLCMSRC can realize better image retrieval and effectively alleviate the semantic gap problem in the sparse space. Compared with the nearest neighbor search algorithm, R GLCMSRC algorithm has a good stable retrieval effect.

关 键 词：图像检索 颜色特征 纹理特征 稀疏分类 内容搜索 特征提取 

分 类 号：TP85[自动化与计算机技术—检测技术与自动化装置]