图像自适应分块单幅超分辨率算法

摘要

在单幅超分辨率图像重建算法中，基于最大后验估计（maximum a posteriori，MAP）算法重建效果和抗噪性能较好，但时空复杂度较高。为了提高模板卷积 MAP（template convolution-based MAP，TC-MAP）算法的运行效率，降低内存消耗，提出了基于图像内容的自适应分块 TC-MAP 新算法，研究了图像分块的最佳尺寸，并根据子块图像的平均梯度，对平滑区域的多个子块进行合并降低分块边界效应的影响，同时采用边界延长进一步抑制分块效应。实验结果表明，算法有效减少了 TC-MAP 算法的运行时间和内存开销，同时保持重建图像质量与原 TC-MAP 算法差别不大。%Single-frame super resolution (SR)image reconstruction algorithms based on maximum a poste-riori (MAP)have preferable reconstruction results and noise robustness but high time and space complexity.To improve efficiency of template convolution-based MAP (TC-MAP)algorithm and decrease its memory consump-tion,a new image-content-based adaptive block TC-MAP algorithm is proposed.Optimal size of image block is studied,and according to average gradients of image blocks,several blocks in smooth areas are combined to re-duce the influence on block boundary effects,further block boundary extension is used to suppress block effects. Experiments show that both run time and memory consumption of TC-MAP algorithms are decreased,while the quality of reconstructed images makes little difference.