基于深度强化学习算法的全视角人脸纹理图像生成方法  

A Method for Generating Full-view Facial Texture Images Based on Deep Reinforcement Learning Algorithms

作　　者：吕周澍 LYU Zhoushu(Zhumadian Vocational and Technical College,Zhumadian,He'nan,China 463000)

机构地区：[1]驻马店职业技术学院,河南驻马店463000

出　　处：《湖南邮电职业技术学院学报》2024年第2期34-38,共5页Journal of Hunan Post and Telecommunication College

基　　金：2021年驻马店市科技攻关项目“基于多源数据的城市暴雨积涝点监测预警关键技术研究”(项目编号:212102210515)。

摘　　要：由于人脸的面部特征复杂且纹理结构多样,传统方法往往受到完整性、纹理真实性、清晰度以及鲁棒性等方面的局限。因此,本研究提出基于深度强化学习算法的全视角人脸纹理图像生成方法。首先,对全视角人脸面部区域进行细致划分,建立坐标系以精确提取各区域的关键纹理结构特征点。随后,将这些特征点输入深度强化学习模型中,通过算法优化整合成一套全面的全视角特征点集合。利用马尔科夫权重场进一步处理特征点,通过计算联合概率,并结合重叠区域约束条件,生成了细节丰富、纹理清晰的全视角人脸纹理图像。实验结果表明,所提出方法生成的图像具有较高的峰值信噪比和较高的纹理清晰度,且鲁棒性较好,有效满足了高质量人脸纹理图像生成的需求。Due to the complexity of facial features and the diversity of texture structures,traditional methods are often limited by integrity,texture authenticity,clarity and robustness.Therefore,a full-view facial texture image generation method based on deep reinforcement learning algorithm is proposed in this study.Firstly,the facial region of the full-view face is divided in detail,and the coordinate system is established to accurately extract the key texture structure feature points of each region.Subsequently,these feature points are input into the deep reinforcement learning model and integrated into a comprehensive set of full-view feature points through algorithmic optimization.By using Markov weight field to further process the feature points,a full-view facial texture image with rich detail and clear texture is generated by calculating the joint probability and combining the overlapping region constraints.The experimental results show that the image generated by the proposed method has higher peak signal-to-noise ratio and higher texture clarity,and is robust,which can effectively meet the needs of high-quality facial texture image generation.

关 键 词：人脸图像生成 人脸纹理图像 深度强化学习算法 图像生成 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]