基于TI-SPCA的人脸自动对齐及识别框架

Unified framework for automatic face alignment and recognition based on TI-SPCA

随着人脸识别算法在众多应用领域的迅猛发展,作为人脸检测和人脸识别中间步骤的人脸对齐算法日益受到重视。针对平面内的人脸图像旋转问题,提出一个基于TI-SPCA(Transformation Invariant Symmetrical Principal Components Analysis)的人脸自动对齐方法及其识别框架。不同于传统的人眼对齐方法,TI-SPCA通过最小化重构图像和扭曲图像之间的误差得到一个旋转不变的特征空间,最终实现无人为干涉的全自动对齐。为了将其性能与人眼对齐方法的性能进行比较,并展示其优势,文中分别在ORL数据库和FERET数据库上通过两种不同对齐方法的输出图像从视觉效果上直观地展现。进一步地,为了验证对齐后的图像在识别算法中的有效性,结合三种距离函数和四种局部算子进行了对比实验,实验结果表明了基于TI-SPCA的全自动对齐方法在人脸识别中的有效性。

With the rapid development of face recognition algorithms in many applications, more and more attentions have been paid to the face alignment algorithm which is the intermediate step between face detection and face recognition. A Transformation Invariant Symmetrical Principal Components Analysis （TI-SPCA） framework is proposed to automatically align and recognize facial images. Different from the traditional eye-aligned methods, TI-SPCA alignment needs no human intervention. Furthermore, TI-SPCA obtains a transform invariant feature space by minimizing the error between reconstructed images and distorted images. To compare its performance against the eyeligned method and show its outperformance, this work intuitively demonstrate the visual performance of alignment through the output images of two different alignment methods on ORL database and FERET database. Finally, in order to verify the validity of the aligned image in the recognition system, the proposed method is evaluated by combining three distance functions and four local operators. The experimental results show the effectiveness of automatic alignment method based on TI-SPCA in face recognition.