Abstract:

The existing Harris corner extraction algorithm in the image matching method has the defects of low accuracy, weak anti-interference and anti-lighting ability. A robust exact matching algorithm based on the combination of Local Binary Patterns（LBP） and Graph Transformation Matching（GTM） is proposed. Firstly, the improved Harris edge feature detection is used to extract feature points and image blocks are selected as feature regions. Secondly, the improved Center Symmetric Local Binary Patterns（CSLBP） are used to reduce the dimensions of high-dimensional features to generate 24-dimensional feature description. Then, coarse matching of images is realized based on Euclidean distance. Finally, graph variation matching method is used to eliminate error matching to improve the accuracy and robustness of matching. The test results show that the proposed algorithm is effective, it not only has good resistance to scale and rotation changes, but also has strong noise suppression and light resistance. The robust LBP proposed in this paper combines the advantages and disadvantages of traditional feature matching algorithms to make the detection and matching results more accurate. Compared with Harris algorithm and SURF algorithm, the stability and accuracy are significantly improved.

Key words: Local Binary Patterns（LBP）, characteristic value, scale invariant feature transform, image matching

摘要：

针对现有的Harris角点提取算法在图像匹配法中，存在精度低、抗干扰和抗光照变化能力弱的缺陷，提出一种基于局部二进制模式（Local Binary Patterns，LBP）和图变换匹配算法（Graph Transformation Matching，GTM）相结合的鲁棒精确匹配算法。采用改进的Harris边缘特征检测提取特征点并选取图像块作为特征区域；采用改进的中心对称局部二进制模式（Center Symmetric Local Binary Patterns，CSLBP）对高维特征进行降维生成24维特征描述子，并依据欧氏距离实现图像粗匹配；采用图变化匹配法剔除误差匹配来改善匹配的精度和鲁棒性。测试结果表明，所建议算法是有效的，它不仅具有良好的抗尺度和旋转变化特性，而且具有较强的噪声抑制能力和抗光照变化能力。提出的鲁棒性算法不仅充分考虑到传统特征匹配算法优缺点，使检测与匹配结果更加准确，而且较Harris算法以及LBP算法稳定性和准确度有了明显的提高。

关键词: 局部二进制模式（LBP）, 特征值, 尺度不变特征变换, 图像匹配