针对传统掌纹识别方法易受噪声干扰, 且旋转鲁棒性差的问题, 提出一种采用均匀局部二元模式 (Uniform Local Binary Patterns, ULBP)及稀疏表示的掌纹识别方法。该方法利用善于表达图像纹理特征, 且具有良好旋转不变性和抗干扰性的 ULBP提取掌纹图像特征; 同时考虑到直接对整幅图像进行 ULBP处理会丢失局部纹理, 采用先对各图像进行分块, 再对各块分别进行 ULBP处理的特征提取方案。在分类算法的设计上, 本文利用掌纹图像库中训练样本的 ULBP特征构造过完备字典, 通过求解 l1范数意义下的最优化问题实现测试样本的稀疏分解, 并提出一种基于统计残差平均的稀疏表示分类方法, 实现了测试掌纹图像的分类识别。实验结果表明, 本文方法不仅具有良好的旋转及噪声鲁棒性, 而且总体识别率明显优于基于 PCA及 2DPCA的传统稀疏表示分类方法, 对于包含 5 000031 0人的掌纹数据库, 识别率分别提高了 8.8%和 6.8%。

In view of the problems of conventional palmprint recognition algorithms that are susceptible to noise interference, and poor robustness to rotation attacks, a novel palmprint recognition method is presented by using Uniform Local Binary Patterns (ULBP) and sparse representation. This method utilizes ULBP which is good at expressing image texture features and has the good rotation invariance and noise immunity characteristics to extract palmprint features. At the same time, taking into account the phenomenon that the local texture features will be lost if we directly extract palmprint features to the whole image by ULBP, this article blocks the palmprint image first and then statistics each sub-block features by ULBP. On the design of sparse classification algorithm, this article takes ULBP features of the training samples to construct a redundant dictionary, and achieves sparse decomposition of testing samples by solving the optimization problem based on l1 norm, and proposes a sparse representation classification method based on statistical average residuals to achieve recognition and classification result. The experiments demonstrate that the proposed method has the good robustness to rotation and noise, and the overall recognition rate is increased obviously. Compared with the traditional PCA and 2DPCA methods, for the database which contains 50 kinds of palmprint images, the recognition rate is increased by 8.8% and 6.8% respectively.