基于能量特征曲面的指关节纹识别方法
首发时间:2020-06-29
摘要:针对目前指关节纹的身份识别方法中存在鲁棒性差的问题,提出一种基于非下采样的Shearlet变换(Nonsubsampled Shearlet Transform,NSST)和Tetrolet变换能量特征的指关节纹识别方法。首先,采用直方图均衡化对图像进行灰度调整,减少因光照分布不均对识别系统产生的影响。其次,利用NSST及其逆变换得到去噪后的重构图像,并对其进行Tetrolet变换,建立低频图像的能量曲面,使其作为指关节纹纹理特征的一种体现。然后,将不同图像的能量曲面作差,得到能量差曲面,进一步计算曲面的方差,并以此为依据对不同指关节纹图像进行分类识别。实验在HKPU-FKP、IIT Delhi-FK、和HKPU-CFK图库及它们的噪声图库中进行,结果显示,本文方法正确识别率可达98.0392%,最低等误率(EER)为2.5646%。相比其他典型和流行算法,本文算法提高了指关节纹识别系统的性能,具有可行性和有效性。
For information in English, please click here
Finger-knuckle-print Recognition Method Based on Energy Feature Surface
Abstract:Aiming at the identification problem with poor robustness based on finger-knuckle-print, a recognition method using nonsubsampled shearlet transform(NSST) and Tetrolet transform energy feature is proposed. Firstly, histogram equalization is used to adjust the gray level of the image to reduce the influence of uneven light distribution on the recognition system. Secondly, NSST and its inverse transform are used to obtain the reconstructed image after denoising, and Tetrolet transform is performed on it to establish the energy surface of low-frequency image, making it a performance of the finger-knuckle-print texture characteristics. Then, the energy difference surface of different images is calculated, and the variance of the surface is further calculated. The experiment was carried out in HKPU-FKP, IIT Delhi-FK, and HKPU-CFK database and their noise database. The results showed that the correct recognition rate of the proposed method reached 98.0392%, with the lowest equal error rate (EER) of 2.5646%. Compared with other typical and popular algorithms, this algorithm improves the performance of the finger-knuckle-print recognition system, which is feasible and effective.
Keywords: Image processing Finger-knuckle-print recognition Energy feature surface
引用
No.****
动态公开评议
共计0人参与
勘误表
基于能量特征曲面的指关节纹识别方法
评论
全部评论0/1000