华中科技大学学报（自然科学版）2024，Vol.52Issue(2)：22-28,7.DOI:10.13245/j.hust.240203
基于三元组网络和混合粒子群的碎纸拼接算法
Splicing algorithm of shredded document based on triplet network and hybrid particle swarm optimization
摘要
Abstract
In order to solve the problems of low clustering accuracy and general splicing effect of heuristic algorithm after fine-grained cutting,an automatic splicing algorithm based on triplet network and hybrid particle swarm optimization was proposed.Splicing of shredded document could be divided into three stages:inter line clustering,compatibility evaluation and intra line splicing.First,in the inter row clustering stage,the projected image vectors were filled adaptively to supplement the missing feature information,and then the Gaussian mixture model clustering algorithm was used for clustering.After that,in the compatibility evaluation stage,with the help of the triplet network model,the clustered images of the same line were mapped to a common metric space,and a distance matrix with obvious discrimination was obtained.Finally,taking this distance matrix as the fitness value of particles,an intra line splicing algorithm based on hybrid particle swarm optimization was proposed.The experimental results show that the algorithm proposed can effectively improve the clustering accuracy and splicing accuracy,which has a good effect on the reconstruction of shredded paper.关键词
碎纸片拼接/混合粒子群算法/高斯混合模型/三元组损失/卷积神经网络Key words
splicing of shredded paper/hybrid particle swarm optimization/Gaussian mixture model/triplet loss/convolutional neural networks分类
信息技术与安全科学引用本文复制引用
陈志刚,苏周,方佳..基于三元组网络和混合粒子群的碎纸拼接算法[J].华中科技大学学报（自然科学版）,2024,52(2):22-28,7.基金项目
科技创新2030-"新一代人工智能"重大项目(2020AAA0109605). （2020AAA0109605）
