基于多光谱图像融合的掌纹识别方法

生物特征识别在信息安全领域发挥着重要作用，掌纹识别作为一种新型生物特征识别方式，具有低失真、非侵入性和高唯一性等优势。传统掌纹研究大多使用自然光成像系统以灰度格式获取，识别精度很难进一步提升。为了获得更多的身份鉴别信息，提出利用多光谱掌纹图像代替自然光掌纹图像。针对现有掌纹识别算法由于没有考虑到不同光谱的特性而导致纹理细节丢失，识别精准率低的问题，提出了一种基于多光谱图像融合的掌纹识别算法。该方法通过对不同光谱下的掌纹图像进行快速自适应二维经验模式分解（FABEMD），将多光谱掌纹图像分解成一系列频率由高到低的二维固有模态函数(BIMF)和一个残余分量，残余分量可被视为该光谱图像低频信息的初步估计。图像采集过程中光照条件很难保持稳定，而近红外光谱图像在进行FABEMD分解时对光照变换敏感，容易导致分解后的BIMF背景信息过于冗余；因此对分解后的近红外掌纹图像进行背景重建及特征细化，在对背景冗余信息进行平滑处理的同时可以有效增强高频信息的特征表达。为避免直接融合处理后引发的图像过度曝光问题，提出对近红外特征压缩后再融合。此外，提出了一种结合了注意力机制的改进残差网络（IRCANet），用于融合后的掌纹图像分类，在网络中引入分阶段残差结构，缓解了网络的退化问题，在学习过程中有效地减少信息丢失，对于融合后的多光谱掌纹图像，分阶段残差结构能够稳定地将图像信息在网络间传输，但对图像中的高低频信息区分效果不够显著，为了使网络关注更多区分性特征，利用特征通道间的相互依赖性，在分阶段残差结构中结合了通道注意力(Channel Attention)机制。最终，在香港理工大学（PolyU）多光谱掌纹数据集上进行的综合实验表明，该方法可以取得良好的效果，算法识别准确率能达到99.67%且具有良好的实时性。

Palmprint Recognition Method Based on Multispectral Image Fusion

Biometric identification plays an important role in the field of information security. As a new biometric identification method, Palmprint identification has the advantages of low distortion, non-invasiveness and high uniqueness. Traditional palmprint research mostly uses natural light imaging systems to acquire in grayscale format, and it is not easy to improve the recognition accuracy further. In order to obtain more identification information, a multispectral palmprint image is proposed to replace the natural light palmprint image. Aiming at the problem that the existing palmprint recognition algorithms do not consider the characteristics of different spectra, resulting in loss of texture details and low recognition accuracy, a palmprint recognition algorithm based on multi-spectral image fusion is proposed. This method decomposes the multispectral palmprint image into a series of two-dimensional intrinsic mode functions (BIMF) with frequencies from high to low and a Residual component, which can be regarded as a preliminary estimate of the low-frequency information of the spectral image. Since the illumination conditions are unstable during the image acquisition process, and the near-infrared spectral image is sensitive to illumination transformation during FABEMD decomposition, it is easy to cause the decomposed BIMF background information to be too redundant. Therefore, the background reconstruction of the decomposed near-infrared palmprint image is performed. And feature refinement, which effectively enhances the feature expression of high-frequency information while smoothing the background redundant information. In order to avoid the problem of image overexposure caused by the spectral information after direct fusion processing, it is proposed to compress the near-infrared features before fusion. In addition, an improved residual network (IRCANet) combined with an attention mechanism is proposed for palmprint image classification after fusion, and a staged residual structure is introduced into the network to alleviate the degradation problem of the network. For the fused multispectral palmprint image, the staged residual structure can stably transmit image information between networks, but the effect of distinguishing high and low-frequency information in the image is not significant enough. In order to make the network pay attention to More discriminative features, use the interdependence between feature channels and incorporate a channel attention mechanism in the staged residual structure. Finally, comprehensive experiments on the multispectral palmprint dataset of the Hong Kong Polytechnic University (PolyU) show that the method can achieve good results, and the algorithm recognition accuracy can reach 99.67% and has good real-time performance.