基于三元组的跨模态深度哈希算法
首发时间:2017-12-04
摘要:跨模态的深度哈希算法可以跨越文本、图像之间的语义鸿沟,使得图片和文字能够相互检索。为了提高现有跨模态哈希算法检索的准确性,本文提出了一个基于三元组的跨模态哈希算法DTCMH(Deep Triplet based Cross Model Hashing),该算法采用生成三元组的训练方式,其损失函数包括减少量化误差的正则项、基于三元组的排序损失以及Softmax分类器中的分类误差损失,更好的学习了数据之间的排序关系以及数据的高级语义特征,最终收到了良好的模型网络训练效果。
For information in English, please click here
Deep Triplet based Cross Model Hashing
Abstract:Cross-modal deep hash algorithm can cross the semantic gap between texts and images so that images and texts can be retrieved from each other. In order to improve the retrieval accuracy of existing cross-modal hash algorithms, this paper proposes a triple-based cross-modal hash model (DTCMH), which loss function takes into account the regular items that reduce the quantization error , the ranking loss based on triplets and the classification error loss in Softmax classifier. This model learns the ordering relationship and the semantic features of data. DTCMH performance is better than the state-of-the-art cross model algorithm.
Keywords: Computer software and theory Deep Hashing Cross Model Triplet
基金:
引用
No.****
同行评议
共计0人参与
勘误表
基于三元组的跨模态深度哈希算法
评论
全部评论0/1000