摘要
针对人脸超分辨率重建中如何获得细节更为丰富的超分辨率重建结果问题,通过评估投影空间的一致性,给出了一种投影空间的选择方法。该方法首先根据图像样本空间与投影空间之间的映射关系计算高低分辨率图像样本所对应投影空间的投影值,然后随机选取若干对高低测试图像样本投影值作为重建目标,并通过邻域嵌入分别获取其对应的高低分重建权值,最后通过计算高低分重建权值间的余弦相似度,并进行直方图统计分析来评估投影空间的一致性并对投影空间进行选择。实验结果表明,该方法可以快速高效地对各投影空间进行评估与选择,其中优秀的投影空间能够将人脸超分辨率重建结果的峰值信噪比提升0.3dB左右。
A selection method of projection space based on assessing the consistency of projection spaces is proposed to achieve more detailed face super-resolution reconstruction results. Firstly , pairs of high and low resolutions are projected into a common space based on the mappingrelationship between the original space and the projection space. Secondly, weights ofreconstruction are calculated through a random selection of projected high and low resolution pairs to obtain cosine similarities. These cosine similarities are then used to evaluate and tochoose projection spaces with histogramming approach. Experiment results show the efficiency of the proposed method. Moreover, with a cautiously selected projection space, the corresponding face super-resolution algorithm achieves about a 0. 3 dB improvement on peak signal to noise ratio.
作者
张哲
齐春
张钊强
陈晓璇
ZHANG Zhe1, QI Chun1, ZHANG Zhaoqian1, CHEN Xiaoxuan2(1. School of Electronics and Information Engineering, Xi＇an Jiaotong University, Xi＇an 710049, China;2. School of Information and Technology, Northwestern University, Xi＇an 710049, Chin)
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2018年第8期43-48,共6页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目(61572395
61675161)
关键词
人脸图像
超分辨率
投影空间
邻域嵌入
face image
super-resolution
projection space
neighbor embedding
