摘要
多尺度局部自相似性是指同一幅图像中存在相同尺度或不同尺度的相似子块,这种图像局部结构自相似性广泛存在于自然图像中。提出了一种基于多尺度局部自相似性结合邻域嵌入的单幅图像超分辨率算法,该算法不依赖于外界图像,仅仅在原始图像的局部子窗口中搜索目标图像块的相似子块,并结合邻域嵌入算法,进一步提高参与重建的图像块与目标图像块的相似性程度。实验结果表明,与双三次插值与传统邻域嵌入算法相比,新算法在保证算法效率的前提下,能有效提升超分辨图像的重建质量。
Multi-scale local self-similarity, which widely occurs in natural images, refers to those similar patches either within the same scale or across different scales coming from the same input image. In this paper, we propose a single image super resolution algorithm based on multi-scale local self-similarity and neighbor embedding; this al- gorithm does not rely on an external example database nor use the whole input image as a source for example pat- ches. Instead, we extract patches from extremely localized regions in the input image and combine with neighbor embedding algorithm, further increasing the similarity between the patches which take part in reconstruction on the one hand, and the target patch on the other. Experimental results and their analysis demonstrate preliminarily that our method can improve the quality of super resolution image as compared with the bicubic interpolation and tradi- tional neighbor embedding algorithm, thus ensuring the efficiency of the algorithm.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2015年第6期1014-1019,共6页
Journal of Northwestern Polytechnical University
基金
国家自然科学基金(61201323)
陕西省自然科学基金(2014JQ5189)资助
关键词
数据库系统
效率
嵌入式软件
误差
实验
滤波器
图像重构
数学运算符
MATLAB
光学分解功率
像素
局部自相似性
多尺度
邻域嵌入
超分辨率
database systems, efficiency, embedded software, errors, experiments, filters, image reconstruction,mathematical operators, MATLAB, optical resolving power, pixels
local self-similarity, multi-scale,neighbor embedding, super resolution
