作　　者： ;

机构地区： 广东工业大学信息工程学院

出　　处： 《微型机与应用》 2017年第16期44-47,52,共5页

摘　　要： 通过对目前存在的一些基于纹理的图像检索算法的分析,指出了它们存在的不足,提出一种基于LBP的加权分布纹理熵的图像检索算法。首先,该算法依据图像关键兴趣区域分布特点将图像分成不均匀的9块子图,接着单独对各个子图块根据Uniform Pattern LBP计算个子图块的纹理特征,并结合信息论统计各个子图的纹理熵,然后把9块子图的纹理熵合并作为整个图像的纹理特征向量,最后依据特征向量元素对应各个兴趣区域的重要性加权进行图像检索。该算法相对传统LBP算法,保留图像空间信息,且区域偏重权系数能弱化不相关信息,同时9维度的特征向量能加快图像的检索效率。实验结果表明,该算法能获得较好的检索性能。 By analyzing the current researches of the texture-based image retrieval algorithm,we point out the defect of each algorithm. This pa-per proposes a newimage retrieval algorithm based on LBP weighting distributed texture entropy. Firstly,based of interest, the image is divided into nine sub-graphs. Then we extract texture features of each sub-graph according to the Uniform Pattern LBPand texture entropy is calculated combined with information theory in each sub-graph. After that the texture entropies of the nine sub-graphs are merged as the texture feature vector of the whole image. Finally,the image vector is weighted according to the importance of the feature vector elements corresponding to each region of interest for image retrieval. Compared with the traditional LBP algoimage space information, and the region weight weakens the irrelevant informat ion. At the same t ime the 9-dimension feature vector acccleratesthe image retrieval efficiency. The experimental results showthat this algorithm can obtain better retrieval performance.

关 键 词： 图像检索 不均匀分布 纹理熵 均匀模式

领　　域： []