改进SURF特征的维吾尔文复杂文档图像匹配检索

针对图像局部特征的词袋模型（Bag-of-Word，BOW）检索研究中聚类中心的不确定性和计算复杂性问题，提出一种由不同种类的距离进行相似程度测量的检索和由匹配点数来检索的方法。这种方法首先需要改进文档图像的SURF特征，有效降低特征提取复杂度；其次，对FAST+SURF特征实现FLANN双向匹配与KD-Tree+BBF匹配，在不同变换条件下验证特征鲁棒性；最后，基于这两种检索方法对已收集整理好的各类维吾尔文文档图像数据库进行检索。实验结果表明:基于距离的相似性度量复杂度次于基于匹配数目的检索，而且两种检索策略都能满足快速、精确查找需求。

Complex Uyghur document image matching and retrieval based on modified SURF feature

This study is aimed at the uncertainty and computational complexity of the clustering center in local image features retrieval based on the bag-of-words (BOW) model. A method to retrieve the measure of similarity degree from different kinds of distance and another method that requires using the matching point number as the basis of retrieval are proposed in this paper. In this method, the SURF feature is first modified to effectively reduce feature extraction complexity, and then FLANN (fast library for approximate nearest neighbors) bidirectional matching and KD-Tree + BBF matching are implemented for FAST + SURF features. Feature robustness is verified under different transformation conditions. Finally, all kinds of Uyghur document images that have been classified and sorted based on these two retrieval methods are retrieved. The results of the retrieval experiments indicate that the similarity degree measure retrieval based on distance is inferior to the retrieval based on matching number, and both of these two retrieval strategies can meet the requirements of fast and accurate searching.