摘要:
针对传统的加速鲁棒性特征(speeded-up robust features,SURF)算法在图像配准中的应用现状,结合图像分块策略和相对距离理论,提出一种基于SURF的图像配准改进算法.通过图像分块策略改善提取特征点分布的均匀性;在SURF算法初匹配基础上,引用相对距离理论剔除异常匹配点,从而提高特征点匹配的精度和可靠性.选取覆盖重庆市沙坪坝实验区的QuickBird卫星数据,以特征点正确匹配率和均方根误差RMSE为量化指标,对所提出的SURF改进算法的图像配准效果进行验证.实验结果表明,改进后的SURF算法的特征点正确匹配率达到88%以上,高于传统SURF算法的76%.通过相对距离剔除误匹配点后,最终配准结果的RMSE达到2.69个像元,符合图像配准的基本需求(RMSE在2个像元左右),具有一定的应用推广价值.
Abstract:
In view of the study status of traditional speeded-up robust features (SURF) algorithm,an improved image registration algorithm based on SURF was proposed in combination with the image blocking strategies and the relative distance theory.The proposed algorithm can improve image uniformity of the feature distribution by image blocking strategy and increase the matching accuracy of the feature point through relative distance theory.With the quantitative indicators of correct feature point matching rate and RMSE,the authors selected the QuickBird satellite data of Shapingba District in Chongqing as the test area to verify the image registration results by using the improved algorithm based on SURF.The results show that the correct feature point matching rate of improved SURF algorithm reached 88%,higher than that of the traditional SURF algorithm (the rate is 76%).Excluding the mismatching points by relative distance,the RMSE of the final registration results reached 2.69 pixels.It meets the basic need of high-precision image registration (the RMSE is 2 pixels around),achieves the automation of remote sensing image registration and thus has some promotional value.
下载: