高级检索
战玉丽, 迟静, 叶亚男, 张彩明, 霍文远. 基于图像相似性和特征组合的超分辨图像重建[J]. 计算机辅助设计与图形学学报, 2019, 31(6): 1018-1029. doi: 10.3724/SP.J.1089.2019.17395
引用本文: 战玉丽, 迟静, 叶亚男, 张彩明, 霍文远. 基于图像相似性和特征组合的超分辨图像重建[J]. 计算机辅助设计与图形学学报, 2019, 31(6): 1018-1029. doi: 10.3724/SP.J.1089.2019.17395
Zhan Yuli, Chi Jing, Ye Yanan, Zhang Caiming, Huo Wenyuan. Super Resolution Image Reconstruction Based on Image Similarity and Feature Combination[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(6): 1018-1029. doi: 10.3724/SP.J.1089.2019.17395
Citation: Zhan Yuli, Chi Jing, Ye Yanan, Zhang Caiming, Huo Wenyuan. Super Resolution Image Reconstruction Based on Image Similarity and Feature Combination[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(6): 1018-1029. doi: 10.3724/SP.J.1089.2019.17395

基于图像相似性和特征组合的超分辨图像重建

Super Resolution Image Reconstruction Based on Image Similarity and Feature Combination

  • 摘要: 针对传统图像重建过程中易丢失细节信息,或在增强细节的同时易产生边缘失真和噪声等问题,提出一种基于图像跨尺度相似性和特征组合的图像超分辨率重建方法.首先利用图像的跨尺度相似性,采用KNN算法分别建立高、低分辨率图像之间的像素特征和梯度特征的映射关系;然后利用像素特征映射关系对输入图像重建包含高频信息的高分辨图像;利用奇异值阈值化获取输入图像的有效高频信息,并利用梯度特征映射关系将高频信息放大后分块叠加到高分辨率图像上,得到最终的图像重建结果.以加州大学图像分割数据库作为实验数据,在Windows7下的Matlab软件进行实验结果展示,实验结果表明,文中方法重建的图像纹理细节丰富、边缘清晰,图像细节显著增强,在视觉效果和客观指标上都有大幅度提升;且该方法无需依赖外部数据库.

     

    Abstract: Aiming at the problems of losing detailed information, or generating edge distortion and noise while enhancing details in traditional image reconstruction process, this paper proposes an image super-resolution reconstruction method based on image cross-scale similarity and feature combination. Using image cross-scale similarity, we first apply KNN algorithm to respectively establish the mapping relationship of pixel features and gradient features between the high and the low resolution images. Then, we use the mapping relation of pixel features to reconstruct the high-resolution image containing high-frequency information from the input image. Finally, we use singular value thresholding to obtain the effective high-frequency information of the input image, and use the gradient feature mapping relationship to enlarge the high-frequency information which will be superimposed onto the high-resolution image to generate the final image reconstruction result. We utilize the image segmentation database of UCLA as the experimental data and display the experimental results in the Matlab software under Windows 7. The experimental results show that our method can reconstruct images with rich texture details, clear edges, and significantly enhanced image details. It greatly improves visual effects and objective indicators. Moreover, our method does not need to rely on external databases.

     

/

返回文章
返回