基于深度学习的K近邻图迭代静脉识别算法研究

Vein recognition algorithm combining K-nearest neighbor and graph iterative based on deep learning

深度学习在计算机视觉中具有强大的特征表达能力,近年来广泛应用于静脉特征的提取与识别。通常,基于深度学习的静脉识别模型在训练阶段,每次仅输入1幅图像及其对应的标签,学习图像与标签之间的映射关系,然而,这种每次只处理单幅图像的方法,难以捕捉不同类别多幅静脉图像之间的关系。为了解决该问题,提出一种基于深度学习的K近邻图迭代静脉识别算法。用较优的深度学习模型提取掌静脉图像特征;利用K近邻算法通过特征距离在训练集中选出最近的K幅图像及其标签,通过这些特征向量生成标签传播矩阵和标签矩阵;利用图迭代算法预测待分类图像的标签,完成分类。在香港理工大学和同济大学提供的掌静脉数据集上进行实验,最高识别精度分别为99.67%和92.72%。

In recent years,deep learning has been widely applied in the extraction and recognition of vein features due to its excellent performance in computer vision.Usually,vein recognition models based on deep learning learn the mapping between a single input image and its label.This approach barely captures the connections between multiple vein images from different categories.To solve this problem,this study introduces a deep learning-based K-nearest neighbor iterative vein recognition algorithm.First,the algorithm extracts features from palm vein images by using advanced deep learning models.Then,it calculates the distances between an image to be classified and training images by using the K-nearest neighbor algorithm,which determines the K most similar images and their labels.A label propagation matrix and a label matrix are created from these feature vectors.Finally,a graph iteration algorithm is used to predict the classifications.Tests are conducted on palm vein datasets provided by Hong Kong Polytechnic University and Tongji University.Recognition accuracies of 99.67% and 92.72% are obtained for the two datasets,respectively.