摘要
针对现有各类非标人像检索存在的问题,进行深入、系统的分析研究,提出一种基于显著性语义属性的交互式非标人像检索方法。通过多标签分类神经网络,生成基于公安行业语义属性的人像表示集;根据人工定义的显著性属性,通过改进的融合相似度距离函数计算方法和快速缩小检索范围的分库权值排序策略实现快速收敛。所提方法通过层进式的交互检索方案成功实施。实验结果表明,在专业用户的操作下,该算法与同类算法相比,将平均检索次数降低到了11.3次,且在查准率、召回率、F1值上均达到了同类算法最优值,很好地解决各类非标人像的检索问题。
Aiming at the nonstandard face retrieval problems,this paper makes an in-depth and systematic analysis and research,and proposes an interactive nonstandard face retrieval method based on saliency semantic attributes.The face representation set based on the semantic attributes of public security industry was generated by multi-label classification neural network.According to the manually defined significance attributes,an improved fusion similarity distance matrix calculation function and the sub base weighted re-rank strategy were adopted to achieve fast convergence.The proposed method was successfully implemented through a layer-by-layer interactive retrieval scheme.The experimental results show that under the operation of professional users,the average retrieval times are reduced to 11.3 times compared with the similar algorithms,and the precision,recall rate and F1 value all reach the optimal value of the same kind of algorithm,which can solve the retrieval problem of all kinds of nonstandard face images.
作者
王茜
陈欣如
Wang Qian;Chen Xinru(Information Center of Criminal Investigation Department,Shanghai Public Security Bureau,Shanghai 200083,China;State Grid Shanghai Shinan Electric Power Supply Company,Shanghai 201199,China)
出处
《计算机应用与软件》
北大核心
2021年第1期205-210,共6页
Computer Applications and Software
