结合显著相似性的局部约束线性编码

发布时间：2018-04-06 00:30

本文选题：图像分类　切入点：局部约束线性编码　出处：《华中科技大学学报(自然科学版)》2017年06期

【摘要】：针对局部约束线性编码(LLC)没有考虑到特征描述符与视觉单词之间的显著相似关系问题,提出了结合显著相似关系的局部约束线性编码.通过显著kNN搜索方法和显著最大池方法将显著相似关系结合到LLC中,首先计算描述符和视觉单词之间的显著相似度,然后分别加入到kNN搜索方法和最大池方法中对LLC进行改进,最后在UIUC8等数据集上进行了实验.该方法相比传统LLC方法及其改进方法,图像分类精度有一定的提升.
[Abstract]:In order to solve the problem of significant similarity between feature descriptors and visual words, LLC (Local constraint Linear coding) is proposed to combine the significant similarity relationship with local constrained Linear coding (LLC).The significant similarity relationship is combined into LLC by significant kNN search method and significant maximum Pool method. The significant similarity between descriptor and visual word is calculated first, and then the LLC is improved by adding kNN search method and maximum Pool method respectively.Finally, experiments are carried out on data sets such as UIUC8.Compared with the traditional LLC method and its improved method, this method improves the accuracy of image classification to a certain extent.
【作者单位】：吉林大学计算机科学与技术学院;中国石油吉林油田分公司扶余采油厂;
【基金】：国家自然科学基金资助项目(61472161,61402195,61502198) 吉林省科技发展计划资助项目(20140101201JC)
【分类号】：TP391.41

【相似文献】