CAJ | 학술논문

针对线性子空间不足以描述头部视角空间非线性变化等因素影响人脸视角流形的精确建模问题,提出一种新的视角流形建模方法,并从理论上将该方法与经典的流形学习建模方法及概念驱动的视角流形建模方法进行比较,通过基于非线性张量分解的人脸及视角识别实验比较视角流形对识别结果的影响,从而给出视角流形的有效性比较.实验结果表明,本文提出的视角流形建模方法比概念驱动的视角流形和TensorFace中的线性视角系数均有更好的识别效果.
침대선성자공간불족이묘술두부시각공간비선성변화등인소영향인검시각류형적정학건모문제,제출일충신적시각류형건모방법,병종이론상장해방법여경전적류형학습건모방법급개념구동적시각류형건모방법진행비교,통과기우비선성장량분해적인검급시각식별실험비교시각류형대식별결과적영향,종이급출시각류형적유효성비교.실험결과표명,본문제출적시각류형건모방법비개념구동적시각류형화TensorFace중적선성시각계수균유경호적식별효과.
Aiming at the issue of linear subspace analysis algorithms being incapable of representing nonlinearity changes in multi-view face images, a novel view manifold modeling method is proposed, which is independent with the identity information. A theoretical comparison among the proposed view manifold, the manifold learning generated one and the concept-driven ones is presented. To verify the validity of the proposed method, the experiments on nonlinear tensor decomposition based identity and view recognition are also given, which show the proposed view manifold achieves better results over the concept driven manifold the linear view coefficients in TensorFace.