基于SURF算法的产品表面缺陷检测研究

Product Surface Defect Detection Based on SURF Algorithm Research

针对产品表面正确性的快速自动无损检测问题,提出了利用垂直投影法确定旋转步长来获取序列图像的方法,并将一种针对尺度旋转不变性(SIFT)改进的SURF算法应用到此方面,该算法通过计算积分图像和Hessian矩阵大大提高了特征点检测的速度,节省了图像匹配时所用的时间,并提高了算法的实时性。首先通过确定旋转步长来获取标准序列图库,其次通过SURF算法寻找最优匹配位置,最后通过相关度的计算来判别各区域是否有缺陷。实验表明,在对待检测图像和标准序列图像库中的5幅图像匹配时SURF算法比SIFT算法大约节省了2.6 s,显然,把SURF算法应用于序列图像中匹配可以大大节省缺陷检测时所用的时间。

Aiming at the fast automatic NDT problem of correctness of product surface, in order to obtain image sequences, a method using vertical projection to determine the rotation is proposed, and the SURF algorithm which includes improved rotating invariance for scale（SIFT）is applied. The algorithm greatly increases the speed of feature point detection by calculating the integral image and Hessian matrix, and saves the image matching time and improves the real-time performance. Firstly, the standard sequence was got by determining the rotating step. Secondly, the optimal matching position was found through the SURF algorithm. Finally, correlation to distinguish whether the regional is flawed was caculated. Experiment results show that the SURF algorithm is betterthan SIFT algorithm, which saves about 2.6 seconds when matching five imagesof the detected image and standard image sequences. It is clear that the SURF algorithm is applied to sequence image matching can greatly save the time.