改进哈希编码加权排序的图像检索算法

Improved Hash code weighted ranking algorithm for image retrieval

针对哈希编码加权排序算法中利用随机采样计算权重,导致权重分配不准确,检索精度较低的问题,提出一种由粗到细的哈希编码加权排序图像检索算法。通过生成较短的哈希编码提升编码效率;利用数据依赖差异得到的采样子集计算哈希码比特位权值,进行加权汉明距离排序得到一个候选最近邻集合;计算集合中数据的得分并重新排序,进一步提高检索精度,实现查询图像的最近邻检索。在手写数字数据集(MNIST)上进行仿真实验,结果表明:当编码长度为48 bit和96 bit时,改进算法的平均准确率可提高13. 33%和11. 61%。

Aiming at the problem that low retrieval precision when using random sampling to calculate weight results in inaccurate weight assignment,a Hash code weighted ranking image retrieval algorithm from coarse to fine is proposed to solve the problem. A shorter Hash code is generated to improve coding efficiency. Weight value of each Hash bit is calculated by the sampling subset obtained by using data dependence difference,and a candidate nearest neighbor set is obtained according to the weighted Hamming distance. Ranking score of the data is calculated in the set,the retrieval precision is further improved by reordering according to the score,and the nearest neighbor search of the query image is realized. The result of the emulation experiment shows that average precision of the proposed algorithm is improved by 13. 33 % and 11. 61 % respectively,when the coding length are48 bit and 96 bit on MNIST database.