摘要: 针对超分辨率图像重建的求解病态性问题,从正则化求解的角度构建数据保真项和正则项,提出一种新的数据融合方法。讨论已有的数据融合方法,利用像素领域和帧间信息控制奇异点,考虑边缘区域的变差权值,避免重建图像的边缘区域过于平滑。实验结果表明,该方法能够提高重建图像质量,具有较好的鲁棒性。
关键词:
超分辨率图像,
病态性问题,
奇异点,
双边全变差,
正则化
Abstract: Super-resolution reconstruction can be considered as a linear system problem which is ill-posed. Aiming at this problem, this paper discusses the construction of the data fidelity term and regularization term in the respect of regularization solution. By comparing the existing data fusion methods, it proposes a method which takes advantage of both the information between pixels in the same frame and neighboring frames to pick out singular points. In order to enhance the performance of Bilateral Total Variation(BTV) filtering on preserving sharp edges, it incorporates a novel implementation which gives the edges heavier variation weighting value and thus achieve better visional effect. Experimental results confirm the superiority of proposed method in both robustness and image quality.
Key words:
super-resolution image,
ill-posed problem,
singular point,
Bilateral Total Variation(BTV),
regularization
中图分类号:
朱高, 王培康, 宋慧慧. 基于变差正则化的超分辨率图像重建[J]. 计算机工程, 2012, 38(17): 235-237,241.
SHU Gao, WANG Pei-Kang, SONG Hui-Hui. Super-resolution Image Reconstruction Based on Variation Regularization[J]. Computer Engineering, 2012, 38(17): 235-237,241.