基于Demons算法的变形掌纹归一化方法研究

Research on the image normalization method of deformed palmprint based on Demons algorithm

非接触采集是主流掌纹采集方式,但其低约束性可能会导致手掌摆放方式不定,与传感器距离不定,从而引起手掌变形,尤其是手掌平面和传感器平面不平行导致的局部变形问题,这将影响后续特征提取,降低识别率。针对此问题,考虑到人手本身是非刚体的特点,提出基于Demons非刚性配准算法的变形掌纹归一化校正模型,进一步增强变形图像与标准图像的相似性,弥补了传统刚性方法校正效果不佳的缺陷。首先使用改进的Demons非刚性配准算法进行变形掌纹的归一化校正,再使用测度指标进行效果评价,结果表明:在任取的图像序列内,与传统的基于归一化互信息(NMI)的刚性配准方法相比,NMI最高提升3.64%,相关系数(COEF)最高提升156.25%,均方误差(MSE)最高降低81.63%,各指标均优于基于NMI的刚性配准方法,验证了本文方法的有效性和优越性,为后续的特征提取和识别创造了有利条件。

Noncontact collection is the main palmprint acquisition mode, but its low restriction may cause different palm placing gestures and different distances between the palm and the sensor. These may result in palm deformation, especially the partial deformation caused by the non parallelity between palm plane and sensor plane; and this may influence subsequent feature extraction and reduce the recognition rate. Considering the non-rigid characteristic of human hands, a normalization model based on Demons non-rigid registration algorithm is proposed to better enhance the similarity between the deformed image and the standard image, and compensate the shortcomings of traditional rigid method that is not very effective. First, the improved Demons algorithm is used to normalize the deformed patmprint; next, the measurement indexes are employed to evaluate the results. Experimental results demonstrate that in randomly selected image sequence, compared with traditional rigid method based on normalized mutual infor- mation （NMI）, the proposed method can increase the NMI by 3.64% , the correlation coefficient （COEF） by 156.25% ,and reduce the mean square error （MSE） by 81.63% at most, which are better than those of the rigid registration method, so the proposed method is effective and superior, and supports favorable conditions for the subsequent feature extraction and recognition.