鲁棒局部二值模式及其在图像检索中的应用

Robust local binary pattern and its application in image retrieval system

针对基于内容的图像检索中对象描述与图像特征间的语义差异以及检索结果易受外界因素影响的问题,提出一种鲁棒局部二值模式(RLBP)结合二进制引力搜索算法(BGSA)的CBIR方法。利用RLBP算子对需要查询的图像和数据库图像进行特征提取;利用BGSA进行特征选择,使特征选择和特征提取并行运行;提取的特征向量与数据库向量进行相似度测度;通过用户的反馈获得系统的偏好,排在前几名的检索图像即最优匹配结果。在Corel、Brodatz和MIT-Vision这3个公开数据库进行4组实验,研究一般情况、有噪声情况、光照变化情况的检索结果,实验结果表明,与同类方法相比,该方法对光照和噪声的鲁棒性更佳,在平均检索率和召回率方面具有一定优势。

Concerning the problems of semantic difference between object description and image feature in content based image retrieval （CBIR） and that retrieval results are easily influenced by external factors, a CBIR method based on robust local binary pattern （RLBP） combined with binary gravity search algorithm （BGSA） was proposed. The RLBP operator was used to extract the features of the image and images of database. The features were selected using BGSA, so that the feature selection and the feature extraction ran in parallel. Similarity measurement was calculated between the extracted feature vectors and database vectors. The preference of the system was obtained through the feedback of the users. Four groups of experiments were carried out in Corel, Brodatz and MIT-Vision public databases, and the retrieval results of general situation, noise and illumination changes were studied. Experimental results show that compared with other similar methods, the proposed method is more robust to illumination and noise, and it has certain advantages in average retrieval rate and recall rate.