摘要
基于稀疏编码的单幅图像超分辨率重建效果较好,但其计算量大、计算复杂,较低的超分速度阻碍了实时应用。为提高超分速度,提出一种基于图像块处理的快速超分辨率重建算法。先将特征提取后的图像块用均值和方差对分成3类,再分别运用不同的方法对图像块进行超分处理,如双三次插值(Bicubic)、邻域嵌入(NE)和稀疏编码超分法(SCSR)等。另外还运用主成分分析法(PCA)对图像块进行降维。实验中与经典的SCSR方法相比较,在获得几乎相同的PSNR值时,超分耗时减少了17.2 s。新方法的超分效果与经典的NE和SCSR方法处于同一水平,超分速度得到了较显著的提高。
The performance of single image super-resolution via sparse coding is promising. But the relatively low speed hinders its real-time application because of the large and complex computation. This paper presents a Proposed Sparse Coding Super-Resolution（PSCSR） algorithm to promote the super-resolution speed. The new method classifies the feature extracted image patches into three classifications with the finite pair of mean and variance. After that, it employs different methods to recover different classified patches, such as Bicubie, NE and SCSR. Additionally, it uses PCA to reduce the patch dimensionalities. In the experiments, the new method is 17.2 s faster than the classic SCSR while the PSNRs are nearly the same. The new method is faster than NE and SCSR with almost the same super-resolution performance.
出处
《电子技术应用》
北大核心
2015年第10期143-146,共4页
Application of Electronic Technique
基金
国家自然科学基金(61271256)
湖北省高等学校优秀中青年科技创新团队计划项目(T201513)
湖北省自然科学基金项目(2015CFB452)
湖北省教育厅科研计划指导性项目(B2015080)
