基于深度学习的金丝猴面部特性的检测与识别算法研究

项目名称： 基于深度学习的金丝猴面部特性的检测与识别算法研究

项目编号： No.61502387

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 许鹏飞

作者单位： 西北大学

项目金额： 19万元

中文摘要： 面向对海量金丝猴图像进行快速处理与分析的需求，展开金丝猴面部图像特征检测与识别的研究。针对面部特征识别中存在的非线性因素的问题，引入深度学习的理论和方法，探索能够学习更本质的面部特征的深层非线性网络结构。内容有：针对金丝猴生活环境的复杂性和不可控制性，利用其面部毛发和皮肤的纹理特征，研究基于特征融合与Adaboost的自然场景图像中面部检测算法；针对面部视角变化引入的非线性因素，利用DBN学习带有视角变化的面部图像之间的非线性联系，提取不同视角下面部图像中存在的共性特征，提出基于DBN的多视角面部识别算法；针对面部表情变化致其面部产生形状和纹理的非线性变化，结合Gabor特征和RBM网络以提取更有效的图像局部结构性特征，并利用面部形变特征，研究基于形变特征和局部结构性特征的面部表情识别算法；针对猴龄估计具有的模糊性，利用DBN学习的面部皮肤的褶皱纹理特征,提出基于模糊DBN的猴龄估计算法。

中文关键词： 金丝猴面部特征；非线性因素；深度学习；面部识别；表情识别

英文摘要： There is an urgent need to process and analyze the massive rhinopithecus images, this project would carry out some relevant research in the field on detection and recognition of rhinopithecus facial features. Due to the problems of nonlinear factors in the process of face recognition, deep learning is introduced to explore a deep nonlinear network structure to learning more intrinsic facial features. And the main research contents include: In view of the complexity and uncontrollability of rhinopithecus' survival environment, using the texture features of their facial hair and skin, and the rhinopithecus' faces are detect based on feature fusion and Adaboost in the natural scene images; For the nonlinear factors introduced by facial angle change, DBN is applied to learn the nonlinear relationship among the multi-view facial images, and to extract the common features，then a multi-view face recognition algorithm based on DBN is put forward; For the nonlinear changes of rhinopithecus' facial shape and texture due to the changes of their facial expressions, the Gabor features and RBM network are combined to extract more effective local structural features, and using the facial deformation characteristics, then a facial expression recognition algorithm is proposed which is based on the deformation characteristics and local structural features; Aimed at the fuzziness of estimating the age of rhinopithecus, DBN is used to learn the fold texture features of their facial skin, and an age estimation algorithm based on fuzzy DBN is put forward.

英文关键词： rhinopithecus facial features; nonlinear factors;deep learning;face recognition;facial expression recognition