基于多分辨率塔式结构的幻觉脸技术的研究  

The research of face hallucination based on multi-resolution pyramid structure

作　　者：郑丽贤[1] 吴炜[1] 杨晓敏[1] 陈默[1] 何小海[1] 

机构地区：[1]四川大学电子信息学院图像信息研究所,四川成都610064

出　　处：《光电子．激光》2008年第9期1244-1249,共6页Journal of Optoelectronics·Laser

基　　金：教育部重点资助项目(107094);四川大学电子信息学院青年教师基金的资助

摘　　要：提出的基于多分辨率塔式结构的幻觉脸算法,使用改进的图像金字塔建立标准人脸训练库作为学习模型,能在多尺度、多分辨率上训练出更有效的先验知识。然后在匹配复原过程中引入一种新思路,先结合塔状父结构搜索出拉普拉斯金字塔中最匹配的4个高频细节,再将这4个高频细节进行加权平均后的结果作为丢失的人脸图像的高频细节,最终复原出超分辨率人脸图像。实验结果表明,该算法对64×64大小的低分辨率人脸图像增强16倍,复原出的256×256大小的超分辨率人脸图像更加平滑,减小了一定的噪声干扰。A novel face hallucination algorithm based on multi-resolution pyramid structure is proposed in this paper. In order to gain more useful multi-scale and multi-resolution prior information,we establish a standard facial training database based on an improved image pyramid as learning model. Then we put forward a new matching algorithm which adopts pyramid parent structure to find the best four match high-resolution patches in the training set. Finally,the average value of the four patches is used to recover the super-resolution face image. The experimental results show that we can enhance a 64×64 face image into a 256×256 one, while the resulting face image appears less noisy and much smoother.

关 键 词：基于学习的超分辨率 幻觉脸 拉普拉斯金字塔 特征金字塔 父结构 

分 类 号：TN911.73[电子电信—通信与信息系统]