基于张量投票的图像超分辨率算法  

Image Super-Resolution Algorithm Base on Tensor Voting

作　　者：胡水祥[1] 黄东军[1] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410083

出　　处：《科技广场》2010年第5期87-90,共4页Science Mosaic

基　　金：国家自然科学基金资助项目(60873188)

摘　　要：张量投票算法在提取图像主观轮廓上具有良好的效果,本文提出了一种基于张量投票的图像超分辨率算法。首先用二维张量矩阵存储低分辨率图像各像素点所处的位置特征信息,并利用稀疏张量投票将特征信息进行加强,再使用稠密张量投票产生高分辨率图像对应的二维张量矩阵,此张量矩阵包含了视觉特性强的边缘信息,最后利用该边缘信息指导高分辨率图像的重构。实验结果表明,该方法得到的高分辨率图像信噪比高、视觉效果好。Tensor voting algorithm has a good effect in extracting subjective contour of image. This article proposes a image super-resolution algorithm based on tensor voting.First of all,using two-dimensional tensor matrix storing each pixel＇s location feature information of low-resolution image, feature information is strengthened by sparse tensor voting. Then a two-dimensional tensor matrix corresponding to high-resolution image is generated by dense tensor voting, the tensor matrix contains edge information with strong visual characteristics. Finally, using the edge information to guide high-resolution image reconstruction. The experimental results show that high-resolution image obtained by our method with high SNR and godd visual effect.

关 键 词：张量投票 超分辨率算法 主观轮廓 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]