基于张量投票的图像超分辨率算法

Image super-resolution algorithm based on tensor voting

张量投票算法在提取图像主观轮廓上具有良好的效果。文章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 paper proposed a image superresolution 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,and 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 good visual effect.