集成局部和全局特征的舰船图像检索算法  

Ship image retrieval algorithm integrating local and global features

作　　者：邢伟寅 罗光明[3] 李礁[2] 钟乐海[2] 韩正勇 XING Wei-yin;LUO Guang-ming;LI Jiao;ZHONG Le-hai;HAN Zheng-yong(Graduate School of JoséRizal University,Manila 1552,Philippines;Mianyang Polytechnic,Mianyang 621000,China;Sichuan Water Conservancy Vocational College,Chongzhou 611200,China)

机构地区：[1]菲律宾黎刹大学研究生院,菲律宾马尼拉1552 [2]绵阳职业技术学院,四川绵阳621000 [3]四川水利职业技术学院,四川崇州611200

出　　处：《舰船科学技术》2023年第5期170-173,共4页Ship Science and Technology

基　　金：四川省科技计划重点研发项目(2022YFG0206);四川省级知识产权专项资金项目(2022-ZS-00156)。

摘　　要：面对现有船舶图像检索存在的“语义鸿沟”问题,导致无法满足用户实际检索需求,研究集成局部和全局特征的船舶图像检索算法,实现不同条件下的船舶图像检索。通过颜色矩方法提取船舶图像全局颜色特征;通过块截断量化编码理念,采用模糊C-均值聚类算法将船舶图像分割成子块,构建子块的二值位图,表示图像局部颜色特征;结合小波变换提取船舶图像纹理特征;求解待检索船舶图像与数据库中船舶图像的各特征相似度,获取总相似度,选取总相似度最大图像作为图像检索输出结果。实验结果表明:该算法可有效提取船舶图像纹理特征;尺度变化、光照变化、旋转条件下的船舶图像检索性能较好,平均匹配率97.94%,平均匹配时间为11.9 ms,检索速度快,操作简单,能够满足用户实时性检索需要。In the face of the semantic gap problem existing in the existing ship image retrieval,which can not meet the actual retrieval needs of users,the ship image retrieval algorithm integrating local and global features is studied to achieve ship image retrieval under different conditions.The global color feature of ship image is extracted by color moment method.Through the concept of block truncation quantization coding,the ship image is divided into sub-blocks using fuzzy C-means clustering algorithm,and the binary bitmap of the sub-blocks is constructed to represent the local color characteristics of the image.The texture feature of ship image is extracted by wavelet transform.Solve the feature similarity between the ship image to be retrieved and the ship image in the database,obtain the total similarity,and select the ship image with the largest total similarity as the ship image retrieval output.The experimental results show that the algorithm can effectively extract the texture features of ship images.The ship image retrieval performance under the conditions of scale change,illumination change and rotation is good.The average matching rate is 97.94%,the average matching time is 11.9 ms,the retrieval speed is fast,and the operation is simple,which can meet the real-time retrieval needs of users.

关 键 词：集成局部 全局特征 船舶图像 检索算法 图像分割 相似度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]