在实际的印刷品缺陷检测过程中, 存在因相机支架的颤动而导致标准印刷图像和待检测图像在空间位置上配准不精确的问题。为此, 在图像去抖动技术的基础上, 提出了一种融合 SURF(speeded-up robust features)和 ORB(oriented FAST and rotated BRIEF)的运动估计算法。首先, 基于 SURF算法提取标准印刷图像和待检测图像的特征点; 其次, 基于 ORB算法对提取的特征点进行描述和匹配; 再次, 将正确匹配的特征点通过仿射模型来求取全局运动矢量; 最后, 通过求得的全局运动矢量来补偿图像, 并完成待检测图像与标准印刷图像的配准。针对待测图像存在的平移、尺度和旋转三种不同变化, 分别采用 SURF-ORB、ORB和 SIFT(scale-invariant feature transform)的运动估计算法进行了性能分析。结果表明, SURF-ORB的特征点匹配对数量最多, 匹配效果最好, SURB-ORB的运动估计时间控制在毫秒级别, 满足现代印刷品缺陷检测的实时性要求。因此, 融合 SURF和 ORB的运动估计算法能够对图像进行精确、实时的配准。

In the actual process of defect detection of printed matter, there is a problem that the standard printed image and the image to be detected are inaccurately registered due to the camera bracket flutter. Therefore, based on the image de-jitter technique, a motion estimation algorithm combining SURF(speeded-up robust features) and ORB(oriented FAST and rotated BRIEF) is proposed. Firstly, the feature points of the standard printed image and image to be detected are extracted based on the SURF algorithm. Secondly, the extracted feature points are described and matched based on the ORB algorithm. Then, the global motion vectors are obtained by affine model of the correctly matched feature points. Finally, the global motion vector is obtained to compensate the image, and the registration between the image to be detected and the standard printed image is completed. In the experiment, the performance of SURF-ORB, ORB and SIFT(scale-invariant feature transform) motion estimation algorithms are analyzed under three different changes. The motion estimation time of SURF-ORB and ORB is controlled at the millisecond level, which meets the real-time requirement of modern printing defect detection. In addition, SURF-ORB has the largest number of feature points matching, and the matching effect is good. Therefore, in the detection process, from the real-time and accuracy of registration, this method in this paper can accurately and real-time registration of images.