基于KCCA的特征融合方法及人耳人脸多模态识别

Feature Fusion Method Based on KCCA and Multi-Modality Recognition Fusing Ear and Face Profile Information

针对非打扰识别问题,鉴于人耳人脸特殊的生理位置关系,提出一种基于二者信息融合的多模态生物特征识别方法.该方法首先采集侧面视角人脸图像,然后将核方法引入到典型相关分析(CCA)中,提出基于核CCA的特征融合方法,并应用其提取人耳人脸的关联特征进行个体的分类识别.仿真实验结果证明了基于KCCA的特征融合方法的有效性.与人耳或侧面人脸单一模态的识别相比,基于人耳人脸的多模态识别的性能显著提高,这为非打扰式生物特征识别提供了一条有效途径.

In order to implement the non-intrusive recognition, a multi-modality recognition method of biometric feature is proposed based on the fusion of ear and face profile information. As there is a special physiological location relationship between ear and eye, only the profile-view face images need to be captured for recognition. By introducing the kernel trick to the canonical correlation analysis （CCA） , a feature fusion method based on kernel CCA （KCCA） is established and is then used to capture the associated feature of ear and eye for classification and recognition. Simulated results indicate that the proposed feature fusion method based on KCCA is effective, and that the multi-modality recognition based on ear and face biometrics is of higher performance than the unimodal bio- metric recognition based on ear or face profile. Thus, there comes an effective approach to the non-intrusive biometric recognition.