一种基于频域的序列图像超分辨率增强方法  

A Sequence Images Super-Resolution Enhancement Approach Based on Frequency-Domain

作　　者：谭政[1,2] 相里斌[1] 吕群波[1] 孙建颖[1] 赵娜[1] 方煜[1] 刘扬阳[1] 

机构地区：[1]中国科学院光电研究院计算光学成像技术重点实验室,北京100094 [2]中国科学院大学,北京100049

出　　处：《光学学报》2017年第7期83-88,共6页Acta Optica Sinica

基　　金：国家自然科学基金(61505219);中国科学院国防科技创新基金(CXJJ-16S045)

摘　　要：针对探测器奈奎斯特频率不足和光学系统点扩展效应造成的成像系统像质下降问题,在频域框架下提出一种改进的多帧图像超分辨率增强方法。在图像序列配准方面,考虑了频谱混叠和插值变换等因素对图像频谱相位差的影响,建立了更准确的亚像元位移提取模型;在图像复原方面,基于同步纹理自回归先验,建立了图像的联合高斯分布模型,并结合贝叶斯方法对插值重建结果进行复原。实验表明,该方法较为全面地考虑了像质下降因素,具有较好的超分辨率增强效果,且计算复杂度较低。Aiming at solving the problem of imaging system degradation caused by the shortage of sensor Nyquist frequency and the point spread effect of optical system,we propose a modified super-resolution enhancement method for multiple input images in the frequency domain framework.In image registration,we build the phase difference model by considering the impact of spectrum aliasing and interpolation transformations,so the sub-pixel information obtained from the model is more accurate.Based on the simultaneous autoregressive prior,the joint Gaussian distribution model combined with the Bayes approach is built to restore the interpolation results.The experimental results demonstrate that the proposed algorithm is effective with low computational complexity.

关 键 词：图像处理 像质下降 超分辨率增强 频域框架 图像配准 图像复原 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]