基于SIFT特征点匹配的印刷品图像检测方法

A Method of Detecting Printed Matter Based on SIFT Interest Points Matching

提出了一种印刷品图像在线检测方法.首先,利用在图像特征点提取领域中运用最为广泛的SIFT(Scale Invariant Feature Transform)算法提取图像稳定特征点,生成特征向量描述符;然后根据向量最近邻(NN)和次近邻(SCN)的距离之比,对匹配点进行初步筛选;最后运用M-estimators法估计特征点对间的几何约束模型,利用该模型进一步精选特征点,确定真正的匹配点对个数,将精选后得到的特征点数与初步筛选得到的特征点数的比值作为判断印刷品是否合格的标准.实验结果表明:该方法能够准确地提取出图像特征点,并通过精选特征点能够在很大程度上改善误匹配问题,从而快速有效地检测出错误的印刷品,得到了良好的检测效果.

A method of detecting defect of printed matter is proposed. Firstly, SIFT feature detector which is widely used is utilized to extract the interest points and generate descriptor of eigenvectors. Then, matching points are filtered by the ratio of vectors＂ distance of the nearest neighbor and that of the second-closest neighbor. Finally, Geometric constraint model of the matching points is estimated with M -estimators, which is used to further filter the matching points. The ratio of the numbers of the twice filtered matching points is the criterion of judging whether the printed matter is acceptable. The experimental results indicate that the method can accurately extract the interest points, and greatly improve the mismatching by accurately selecting the matching points. The defects can be detected efficiently and good effects can be obtained.