摘要
首先将低分辨率及相应高分辨率图像进行非下采样Contourlet变换得到细节图像对,然后通过多任务学习估计细节图像之间的映射参数对低分辨率图像进行重建。实验结果表明,该方法能够取得比传统线性插值、双三次方法更好的效果。
The nonsubsampled contourlet transform is an analytical tool for image with the properties of multi-resolution, multi-direction and translation invariance. Multi-task learning, also called multi- output learning, is of high precision and good anti-noice capability based on the consideration of the relationship among tasks. The low resolution image and the corresponding high resolution image were firstly decomposed into nonsubsampled eontourlet transform domain. Then the map parameters were estimated by multi-task learning. By using these parameters, any low resolution image can be recon- structed. The experimental results demonstrate that the proposed method performs better than bilinear, bicubic interpolation.
出处
《海军工程大学学报》
CAS
北大核心
2009年第1期68-72,共5页
Journal of Naval University of Engineering
