Abstract: In order to improve recognition performance and solve security issue of palm identity verification system, this paper proposes adaptive threshold cancelable palmprint templates based on Local Gabor Directional Pattern with adaptive threshold coding method by mean （mLGDP）, Difference Local Gabor Directional Pattern with adaptive threshold coding method by mean （mDLGDP） and feature fusion of them. After block processing the coding image, feature and binary vector are extracted. Then Bloom filter is used for many to one mapping and palmprint image position scrambling. Finally, cancelable template is generated by using a user key matrix convoluted with final image, which achieves irreversibly transformation. Both theoretical analysis and experimental results show that, even when the key is lost, using two kinds of improved methods respectively can still maintain a high recognition rate, when two kinds of feature are integrated, the recognition rate can effectively improve and achieve better security.

Key words: Gabor, Local Directional Pattern（LDP）, adaptive threshold, feature integration, palmprint, cancelable template

摘要： 针对掌纹身份认证中存在着识别率和安全性较差的问题，提出一种基于多方向的Gabor滤波和局部方向模式（Local Directional Pattern，LDP）的自适应阈值特征编码方法mLGDP，在此基础上，进一步提出一种基于多方向Gabor滤波和LDP方法的自适应阈值差值特征编码方法mDLGDP，并将这两种方法的特征相融合，有效增强了原有掌纹模板间的多样性和识别率。通过对图像的特征编码进行分块处理，提取特征向量并二值化，再采用Bloom滤波器实现多对一映射和对掌纹图像的位置置乱，将得到置乱结果矩阵和用户密钥通过卷积运算进行不可逆变换，最终获得掌纹图像的可撤销模板。理论分析和实验表明，即使在密钥丢失时，分别使用两种改进方法依然可以保持较高的识别率，当使用两种特征相融合的方法时，识别率能够得到有效提高，且具有更好的安全性。

关键词: Gabor滤波, 局部方向模式（LDP）, 自适应阈值, 特征融合, 掌纹, 可撤销模板