摘要： 使用SIFT算法对文字图像进行特征提取时，产生的特征点数目较少，且不同文字产生的特征向量存在强干扰性，导致匹配准确率较低。为此，提出一种改进的SIFT算法。该算法利用二值化图像代替灰度图像，增加特征点数目，并取消SIFT的旋转不变性。实验结果证明，与标准SIFT算法相比，改进SIFT算法能有效提高文字图像匹配的准确率。
关键词:
文字图像,
模板匹配,
尺度不变特征变换算法,
极值点,
特征向量,
特征提取
Abstract: Due to the particularity of the text images, using the Scale Invariant Feature Transform(SIFT) algorithm to extract can not get enough feature points. Otherwise, strong inferences between different words lead the dad matching result. Aiming at this problem, this paper makes a research on the SIFT algorithm and proposes an improved SIFT algorithm for text images template matching. The improved algorithm uses three methods to improve the efficiency, such as to use threshold images instead of gray ones. It increases the number of feature points and cancels the rotational invariance. Experimental result shows that the improved algorithm can effectively improve the accuracy of the text template matching than standard SIFT algorithm.
Key words:
text image,
template matching,
Scale Invariant Feature Transform(SIFT) algorithm,
extreme point,
feature vector,
feature extraction
中图分类号:
胡海青, 谭建龙, 朱亚涛, 龚国成, 刘金刚. 改进SIFT算法在文字图像匹配中的应用[J]. 计算机工程, 2013, 39(1): 239-243.
HU Hai-Jing, TAN Jian-Long, SHU E-Chao, GONG Guo-Cheng, LIU Jin-Gang. Application of Improved SIFT Algorithm in Text Image Matching[J]. Computer Engineering, 2013, 39(1): 239-243.