基于方向局部极值模式的图像检索算法  

Image retrieval algorithm based on directional local extreme value model

作　　者：卫丽华 朱鹏程[2] 管致锦[3] 

机构地区：[1]南通理工学院软件工程系,江苏南通226002 [2]南通大学杏林学院,江苏南通226019 [3]南通大学计算机科学与技术学院,江苏南通226019

出　　处：《计算机工程与设计》2016年第12期3334-3339,共6页Computer Engineering and Design

基　　金：江苏省高校自然科学研究面上基金项目(16KJB520039);南通理工学院科研课题基金项目(2014002)

摘　　要：针对基于局部二值模式(local binary patterns,LBP)的图像检索算法对局部噪声敏感,导致算法抗噪能力弱、鲁棒性低等不足,提出一种基于方向局部极值模式(directional local extrema patterns,DLEP)的图像检索算法。通过计算图像像素点在0°、45°、90°和135°这个方向的局部极值,得到这4个方向的DLEP模式,利用DLEP在纹理提取过程中获取更多的方向边缘信息和空间关系;基于4个方向的DLEP模式,形成图像方向的DLEP直方图,根据DLEP直方图,构建图像的特征向量;引入相似距离度量进行特征匹配,完成图像检索。实验结果表明,与LBP、基于块的LBP算法和中心对称局部二值模式进行比较,在Corel图像数据库中,该算法具有更高的查准率、查全率和平均检索率。To solve the defects such as weak ability of anti-noise,low robustness and sensitivity to the local noise in current image retrieval algorithm based on local binary patterns（LBP）,an image retrieval algorithm based on directional local extrema patterns was proposed.The DLEP model of four directions was obtained by calculating the local extremum in four directions of 0degrees,45 degrees,90degrees,4degrees and 135 degrees,and the directional edge information and spatial relations were got using DLEP during the process of texture extraction.The DLEP histogram of the image orientation was formed based on the four directions of the DLEP model for constructing the feature vector according to the DLEP histogram.The image retrieval was finished by introducing the similarity distance measure.Experimental results show that this algorithm has higher precision,recall and average retrieval rate compared with LBP,block-based LBP algorithm and center symmetric local binary model.

关 键 词：图像检索 局部二值模式 方向特征 相似距离度量 方向局部极值模式 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]