改进的SURF算法在图像匹配中的应用

Application of improved SURF algorithm in image matching

针对加速稳健特征(Speeded Up Robust Features,SURF)算法在三维重建中匹配准确率低的问题,提出基于SURF算法的改进算法。首先利用SURF算法提取特征点,通过近邻搜索(Best Bin Fast,BBF)算法实现Kd-Tree快速查找最近邻特征点,结合双向唯一性匹配的方法完成图像匹配,然后在视差约束下,利用视差梯度约束对初始特征匹配对进行预处理,筛选掉一些偏差较大的匹配对,最后采用随机抽样一致(Random Sample Consensus,RANSAC)算法对特征点二次优化和去噪处理。将其他改进算法和提出的改进算法分别进行图像匹配处理比较,分析算法的性能,得到提出的改进算法匹配成功率达96.3%。实验结果证明提出的改进算法简单快速,匹配精度高。

An improved algorithm based on SURF algorithm is proposed to improve the matching accuracy of SURF(speeded up robust features) algorithm in three-dimensional reconstruction. The feature points are extracted by means of the SURF algorithm,Kd-Tree searches the nearest neighbor feature points quickly by means of BBF(best bin fast)algorithm,and then the image matching is completed in combination with the biuniqueness matching method. Under the disparity constraint,the initial feature matching pairs are preprocessed by means of the disparity gradient constraint,some matching pairs with large deviation are screened out,and the quadratic optimization and de-noising processing of feature points are carried out by means of RANSAC(random sample consensus) algorithm. The performance of other improved algorithms and the proposed improved algorithm were compared and analyzed respectively for image matching,from which the matching success rate of the proposed improved algorithm is 96.3%. The experimental results show that the proposed improved algorithm is simple,fast and has high matching accuracy.