局部特征的局部敏感哈希专利二值化图像检索  

Patent Binary Image Retrieval Method Based On Local Features and Locality Sensitive Hashing

作　　者：董小灵[1] DONG Xiaoling(Patent Literature Department,China National Intellectual Property Administration Patent Office,Beijing 100088,China)

机构地区：[1]国家知识产权局专利局专利文献部,北京100083

出　　处：《电视技术》2022年第5期54-60,66,共8页Video Engineering

摘　　要：基于内容的专利二值化图像检索实现过程,其核心问题是图像内容特征的提取表示和匹配度量。图像内容特征的提取表示通过利用尺度不变特征转换(Scale Invariant Feature Transform,SIFT)算法提取专利二值化图像的局部特征,然后基于旋转90°、旋转180°、尺寸扩大2倍以及尺寸缩小1/2等四种几何变换计算减少局部特征数量,实现对专利二值化图像内容全面、准确的描述表示。图像内容特征的匹配度量通过对专利二值化图像的SIFT特征矢量进行二值化处理并且运用局部敏感哈希(Locality Sensitive Hashing,LSH)方法建立图像特征矢量的索引结构,降低专利二值化图像SIFT特征矢量的匹配复杂度,提升专利二值化图像SIFT特征矢量的匹配效率。本文利用不同规模的专利二值化图像数据集合进行上述检索算法的实验测试,结果表明,该检索算法能够实现专利二值化图像较好的检索速度与查全率。In the implementation of content-based patent binary image retrieval,the core problem is the extraction,representation and matching measurement of image content features.The extraction and representation of image content features extracts the local features of the patent binarized image by using the Scale Invariant Feature Transform(SIFT)algorithm,and then calculates and reduces the number of local features based on four geometric transformations:90 degree rotation,180 degree rotation,2-fold size expansion and 1/2size reduction,so as to achieve a comprehensive and accurate description and representation of the content of the patent binarized image.The matching measurement of image content features is to binarize the SIFT feature vector of the patent binarized image,and establish the index structure of the image feature vector by using the Locality Sensitive Hashing(LSH)method,so as to reduce the matching complexity of the SIFT feature vector of the patent binarized image and improve the matching efficiency of the SIFT feature vector of the patent binarized image.In this paper,the above retrieval algorithm is tested by using patent binary image data sets of different sizes.The results show that the retrieval algorithm can achieve better retrieval speed and recall of patent binary image.

关 键 词：专利二值化图像 局部特征 尺度不变特征转换(SIFT) 局部敏感哈希(LSH) 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]