改进ORB特征点检测的图像拼接方法

Image Mosaic Method Based on Improved OBR Feature Points Detection

针对图像拼接的实时性、准确度问题,本文提出了一种基于改进ORB特征点检测的图像拼接方法.方法在提取图像噪声点周边点的方式和判断特征点条件上进行改进.首先根据ORB算法判断特征点的特点将选取16个点减少为仅选取上下左右4个点,大大减少了质心与周边像素的比较次数;然后根据特征点延展性的特性,再在外圆选取4个相关点进行比较,快速区分噪点或是特征点,提高了特征点匹配准确率.本文还改进了随机取样一致性RANSAC算法,在迭代过程中增加合适的新特征点,提高内点比,减少迭代次数,不仅提升了效率,还可以计算出更合适的变换矩阵,达到更好的拼接效果.实验证明通过改进的ORB算法,改进的RANSAC算法和泊松融合算法能够实时的、高效的、准确的实现图像拼接.

Aiming at the problem of real-time and accuracy of image mosaic, this paper proposes an image mosaic method based on improved ORB feature point detection.The method is improved in the way of extracting the surrounding points of image noise points and judging the condition of feature points.Firstly, according to the characteristics of ORB algorithm, 16 points are selected to select only 4 points, which greatly reduces the number of comparison between centroid and peripheral pixels.Then, according to the characteristics of extensibility of feature points, four related points are selected in the outer circle for comparison, which can quickly distinguish noise points or feature points, and improve the matching accuracy of feature points.This paper also improves the random sampling consistent RANSAC(Random Sample Consensus)algorithm, adding appropriate new feature points in the iterative process, improving the internal point ratio, reducing the number of iterations, which not only improves the efficiency, but also can calculate a more suitable transformation matrix to achieve better splicing effect.Experiments show that the improved ORB algorithm, improved RANSAC algorithm and Poisson fusion algorithm can achieve real-time, efficient and accurate image mosaic.