结合哈希网络和敏感散列的图像检索推荐研究-《计算机技术与发展》

文章信息/Info

Title:: Research on Image Retrieval Recommendation Based on Hash Network and Sensitive Hashing

Author(s):: PAN Hua-feng; WANG Chun-ling; WU Tao; School of Computer Science,Xi’an Polytechnic University,Xi’an 710600,China

Keywords:: image retrieval recommendation; hash network; spherical hash coding; local sensitive hashing; swimsuit version image

摘要:: 面对当今社会的各种海量图像数据,基于图像内容的检索方法对于检索结果的查全率和查准率较为差强人意,并且对于相似图像的检索也会花费较长的时间。为了提升检索效率和检索结果的准确性,提出一种结合深度哈希网络和局部敏感散列的检索推荐方法。首先建立深度哈希网络模型完成对于图像内容特征的提取,并利用球哈希编码优化计算得到汉明空间距离作为特征度量方式,根据度量结果使用局部敏感散列构建索引表提高检索效率;然后对于被检索目标图像进行特性提取,计算汉明空间距离完成特征度量和散列映射,最后可以在索引表中匹配到最相似的若干图像,作为检索到的推荐图像。以泳装版型图像进行实验测试,所构建的推荐模型可以较为快速地完成相似图像的检索,具有较高的准确率。实验结果表明,设计的检索推荐方法基本可以实现相似图像的高效检索。

Abstract:: Facing all kinds of massive image data in today’s society,the retrieval method based on image content is not satisfactory for the recall and precision of the retrieval results,and it will take a long time for the retrieval of similar images. In order to improve the retrieval efficiency and the accuracy of retrieval results,a retrieval recommendation method combining deep hash network and local sensitive hash is proposed. Firstly,the deep hash network model is established to extract the features of the image content,and the Hamming space distance is obtained by using the spherical hashing coding optimization calculation as the feature measurement method. According to the measurement results,the local sensitive hashing is used to construct the index table to improve the retrieval efficiency. Then,feature extraction is carried out for? the retrieved target image,and the Hamming space distance is calculated to complete feature measurement and hash mapping. Finally,the most similar images can be matched in the index table as the recommended images retrieved. Through the experimental test with swimsuit version images,the recommendation model constructed can complete the retrieval of similar images relatively quickly with a high accuracy. The experimental results show that the proposed retrieval recommendation method can basically achieve efficient retrieval of similar images.