项目名称: 人脸美丽智能感知及其应用
项目编号: No.61472144
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 计算机科学学科
项目作者: 金连文
作者单位: 华南理工大学
项目金额: 84万元
中文摘要: 人脸美是人类所广泛感知的一个概念,项目针对人脸美丽智能感知的科学问题,从人脸美学的特征提取、特征子空间流形学习、分类预测、人脸美化等四个方面来探索美丽的本质,主要研究内容包括:(1)人脸美丽的特征提取方法的研究,基于鉴别局部对准(DLA)流形学习模型、以及基于鉴别信息保留(DIP)流形学习模型的人脸美丽子空间流形学习方法的研究;融合DLA及OSS相似性度量学习的人脸美丽度量学习方法的研究;(2)基于深度学习的人脸美丽的特征学习及分类方法的研究,包括稀疏自编码机(SAE)的人脸美丽特征学习方法、基于深度卷积神经网络的人脸美丽分类与预测模型;(3)基于边缘保持的人脸图层分解技术、多层次自由变形技术(MFFD)、平均脸数据驱动、以及人脸纹理合成技术的人脸美化方法的研究;(4)人脸美丽智能感知应用系统。通过本项目研究,预期能使得计算机在某种程度上也拥有像人一样的美丽感知智能。
中文关键词: 图像分析;特征提取;人工智能;图像增强
英文摘要: Facial beauty is an universal part of human experience. This project aims to investigate the fundamental problems of automatic perception of human facial beauty in terms of feature extraction, feature subspace manifold learning, classification, prediction and facial beautification. The main research issues includes, (1) feature extraction approach for facial beauty representation, feature subspace learning methods based on Discriminative Locality Alignment (DLA) manifold learning, Discriminative Information Preservation (DIP) manifold learning; and beauty similarity metric measure using DLA-OSS model (2) facial beauty classification and regression methods based on deep learning theory and technologies, such as the Sparse Auto-Encoder (SAE), deep Convolutional Neural Networks (CNN); (3) face beautification method based on edge-preserving smoothing, Multilevel Free-form Deformation (MFFD), data driven by average faces and facial texture synthesis; (4) design and implementation of a real time facial beautification application system. It is expect that the machine can also be trained with the ability to perception of facial beauty like human beings.
英文关键词: Image Aanlysis;Feature Extraction;Artificial Intelligence;Image Enhancement