基于PLS小波双立方配比插值的图像复原研究  

Image Restoration Based on PLS and Wavelet Bi-cubic Interpolation

作　　者：田洲[1] 孙玉秋[1] 谢先招 张丽玲[1] 邢耀 

机构地区：[1]长江大学信息与数学学院,湖北荆州434023

出　　处：《长江大学学报（自科版）（上旬）》2013年第11期66-69,6,共5页JOURNAL OF YANGTZE UNIVERSITY (NATURAL SCIENCE EDITION) SCI ＆ ENG

基　　金：国家自然科学基金资助项目(60572048);国家大学生创新实验计划项目(201210489333)

摘　　要：针对人脸图像复原问题,提出了基于偏最小二乘法回归(Partial Least Squares,PLS)的超分辨率复原技术与小波双立方配比插值放大结合的算法。对低分辨率图像进行小波分解后得到的3个高频子图像插值,然后用小波逆变换进行重建。同时对低分辨图像进行双立方插值,插值后的图像与小波重建图像进行融合,提取融合图像特征作为训练样本,应用偏最小二乘法回归进行样本训练,最后得到复原高分辨率图像。试验结果表明,基于偏最小二乘法小波双立方配比插值的图像复原算法比传统插值算法(最近邻插值、双线性插值、三次样条插值)在人脸图像复原方面效果好。For human face image restoration problem,a novel super-resolution algorithm is put forward based on Partial Least Squares(PLS)regression and a wavelet bi-cubic ratio interpolation algorithm.The low resolution image is decomposed into three high-frequency images and a low-frequency image by wavelet transform.Three high-frequency images are interpolated before reconstruction by wavelet inverse transform.At the same time,the original low resolution image is interpolated by bi-cubic interpolation.The interpolated image and reconstructed image by wavelet transform are fused.The characteristics are abstracted from fused images as trained samples.The high resolution image is restored according to partial least squares regression.It is shown from experimental results that the super-resolution restoration algorithm based on partial least squares and wavelet bi-cubic ratio interpolation algorithm is better than the other traditional algorithms.

关 键 词：超分辨率 偏最小二乘法 小波双立方插值 图像复原 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]