一种基于角点特征的图像自动配准方法

An Automated Image Registration Method Based on Corner Feature

图像配准是图像处理和计算机视觉中的重要环节。提出了一种基于角点特征的图像自动配准方法来处理具有相似变换的图像配准问题。角点特征由改进的Harris算子提取,然后将提取的角点组成虚拟三角形,利用在相似变换下参考图像和待配准图像中对应的虚拟三角形相似的原理,找到最相似的两个虚拟三角形,以它们对应的顶点作为控制点,求出变换模型参数,从而配准两幅图像。该方法只要求两幅图像中提取的角点特征包含3个以上的对应角点,就能配准两幅图像。它的另一个优点是理论上对两幅图像之间发生的平移、旋转和尺度变化没有限制。实验结果表明:这种图像自动配准算法是正确和有效的。

Image registration plays an import role in image processing and computer vision. An automated image registration method based on corner feature is proposed to deal with the registration of image of similarity geometric distortion in this paper. Corners are extracted by an improved technique based on Harris and then compose virtual triangles. According to the principle that the corresponding virtual triangles are similar in the reference and sensed images under the similarity transformation, the most similar two virtual triangles can be found. Their corresponding vertexes are used as control points and the parameters of similarity transformation are obtained. The proposed method is guaranteed to register images if only three corresponding corners are extracted from the reference and sensed images. Another advantage of the method is that there is theoretically no limited to scale, translation and rotation of two images. The experiment results prove that the proposed method is accurate and effective.