一种通用形状的印鉴配准算法应用

An Image Registration Method for Chinese Seal Identification

提出了具有各形状通用性的包括粗略配准及精确配准的双层印鉴配准方法。在粗略配准层中,利用极坐标变换将旋转操作变换为平移操作,可避免在直角坐标系中图像旋转时的插值误差的影响;在极坐标系中,用投影卷积的方法可以提高配准时对印文内容不一致性的鲁棒性。在精确配准层中,通过启发式搜索完成位置的精确配准。实验表明,本方法具有较好的通用性,同时可获得高准确度的配准结果。

This paper presents a double-level registration method which includes a rough registration and a precise registration. In the rough registration level, a seal image is transformed form rectangular coordinates into polar coordinates. So rotation in rectangular coordinates is transformed into translation in polar coordinates. Then projection convolution in polar coordinates is used to improve robustness to dissimilarity of seal impressions during seal registration. In the precise registration level, a heuristic search is used to complement the precise registration of translation. Experiments show that high registration results can be obtained by the proposed method.