随机维哈希量化视词字典的目标检索方法

OBJECT RETRIEVAL BASED ON VISUAL-WORDS DICTIONARY WITH RANDOMISED DIMENSIONS HASHING QUANTIFICATION

基于视词字典树的算法由于高效性使其在基于大规模图像数据库的目标检索领域得到了广泛地应用。该类算法属于从文字搜索领域借鉴来的"视觉词袋"的算法。这种算法中的一个关键步骤是将高维特征向量量化成视词。将这种量化过程看作高维特征向量的最近邻搜索问题,并且提出一种随机维哈希(RDH)算法用于索引视词字典。实验结果证明,该算法比基于字典树的算法具有更高的量化精度,从而可以显著提高目标检索性能。

Visual-words dictionary tree-based algorithm has been widely applied in object retrieval in large-scale image database due to its efficiency. Such algorithm appertains to the bag-of-visual-words algorithm which is borrowed from text search field. A key step of such algo- rithm is to quantify the high-dimensional feature vectors to the visual words. In this paper, we consider the quantification process as the nea- rest neighbour search of high-dimensional feature vectors, and propose a randomised dimensions hashing algorithm to index the visual-word dictionary. Experimental results demonstrate that the proposed algorithm has higher quantification accuracy than the vocabulary tree-based al- gorithms, thus it can significantly improve object retrieval performance.