一种具有平衡标签预测能力的在线哈希算法

Online hash algorithm with balanced label prediction ability

针对传统离线哈希算法训练模型耗时、占用内存大和不易更新模型的问题,以及现实图像集的标签存在大量损失的现象,提出了一种能够平衡标签预测的在线哈希算法(BLPOH)。BLPOH通过标签预测模块生成预测标签,并融合残缺的真实标签,能够有效缓解因标签损失导致的模型性能下降。观察到标签存在分布不平衡现象,提出标签类别相似性平衡算法并应用于标签预测模块,提升标签预测的准确性。将旧数据的信息加入哈希函数的在线更新过程,提升模型对旧数据的兼容性。通过在两个广泛使用的数据集上进行实验,并和一些当前先进的算法进行对比,结果证实了BLPOH的优越性。

Aiming at the problems that the traditional offline hashing algorithm is time-consuming to train the model,occupies a large amount of memory and is difficult to update the model,and there is a large loss of labels in real image sets,this paper proposed a balanced label prediction for online hashing algorithm(BLPOH).BLPOH generated predicted labels through the label prediction module,and fused the incomplete real labels,which could effectively alleviate the performance degradation of the model caused by label loss.It observed that there was an imbalance in the distribution of labels,then proposed a label category similarity balance algorithm to apply to the label prediction module to improve the accuracy of label prediction.It added the information of old data to the online update process of the hash function to improve the compatibility of the model to the old data.By conducting experiments on two widely used datasets and comparing with some state-of-the-art algorithms,the results demonstrate the superiority of BLPOH.