融合2D-PCA及稀疏表示的掌纹识别方法

提出一种基于稀疏表示的掌纹识别方法,该方法借鉴二维主成分分析(PCA)良好的数据压缩属性和较快的特征提取速度,生成掌纹特征图像.二维PCA不仅克服了一维PCA数据维数过大不易计算的缺点,而且保留了原始图像的数据结构,提取的特征能更好的代表原始图像.为了便于稀疏表达,对提取的掌纹特征图像利用一维主成分分析进行二次特征提取,得到训练样本.虽然此处使用了一维PCA,但是由于这是二次特征提取,提取的特征还是保留了原始图像的数据结构,相比单纯的一维PCA,提高了识别率.利用训练样本构造出冗余字典,并采用稀疏表示理论将测试样本表示为字典原子的线性组合,然后根据表示系数的稀疏性与稀疏集中度实现分类识别.由于该方法利用了表达系数的稀疏性,因此减小了算法的时间和空间复杂度.实验表明,针对香港理工大学的MSpalmprints Database,本文方法的识别率较传统方法有明显提高.

A Method of Palmprint Recognition Integrated by 2D-PCA and Sparse Representation

A palmprint recognition method is presented based on sparse representation, which takes advantage of two-dimensional principal component analysis (2D-PCA) of its better data compression property and faster feature extraction speed to generate the palmprint feature image. The 2D-PCA method not only overcomes the shortage of complex calculation by PCA method due to its higher data dimension, but also retains the data structure of original image to obtain better feature. In order to facilitate sparse representation, we take the PCA method to extract features of palmprint feature image to obtain the training samples. In this case, the training samples still retain the data structure of original image and improve the recognition rate compared to simple PCA method. We take the training samples to construct a redundant dictionary, and express the testing samples as a linear combination of dictionary atoms by sparse representation theory. Then, the classification is achieved according to the sparsity of representation coefficient and sparse concentration. Due to the sparsity of representation coefficient, this method reduces the time and space complexity. Andthe experiments show that the recognition rate of this method is obviously higher than traditional method for the Hong Kong Polytechnic University MSpalmprints Database.