中文摘要：

针对传统的位置敏感哈希编码低效的问题,提出一种监督学习框架下基于正交子空间的判别投影哈希函数学习的海明编码方法.该方法首先根据特征值的能量分布进行子空间分解,其次基于Fisher判别分析准则,利用样本的分布信息学习一组最佳投影的哈希函数,实现原始特征空间向海明空间的紧致嵌入,最终生成一组紧凑且具有判别性的二进制编码,并用于图像检索.在公开数据集上的实验结果表明：该算法与其他经典算法相比,具有较好的稳定性,降低了内存消耗并提高了检索的平均准确率.

英文摘要：

We propose a supervised discriminative hash function learning for hamming coding to deal with the inefficiency of classical locality sensitive hashing（ LSH） method. Firstly,the method decomposes the subspace according to the energy distribution based on the eigenvalues. Secondly,it makes use of the distribution of samples based on fisher criterion to embed the original feature space into hamming space in a more compact manner,and finally it generated a compact and discriminative binary code for the image retrieval system. Experimental results based on the public dataset demonstrated that our method is superior to other classical methods,which is more stable with less memory cost and also increases the average precision.