一种基于MAP的图像超分辨率重建算法

引入一种基于关键点滤波(Critical-Point Filters,CPF)的图像配准方法,并在最大后验概率(Maximum a Posteriori,MAP)框架下提出一种改进的集投影法(Projections onto ConvexSets,MAP/POCS)混合算法。算法把POCS的残差约束集合加入到基于CPF图像配准的MAP正则算法中,在每次迭代重建中对重建图像的像素点进行约束,充分利用这三种算法的优点。实验结果表明,相比于传统的重建方法,该算法能够更有效地表达视频中的非平移运动,超分辨图像主观质量有明显改善。

MAP-based of super-resolution reconstruction algorithm

Super-resolution reconstruction uses existing imaging system, to improve the resolution reconstruction by the image signal processing. Traditional SRR adopts block matching, optical flowand so on.It has some defects such as low matching accuracy and sensitive to the variety of the light so that it can noteffectivelyapply to the super-resolution reconstruction on the images fromvideos.For expressing the non-translational motion from videos effectively, this paper introduces an image registration method based on the critical-point filters, and proposes animproved MAP/POCS hybrid algorithm based on the framework of MAP.In this paper,distinct fromtraditional MAP/POCS hybrid algorithm which was simply serially combining the two algorithm independently,the POCS residual set of constraints were added to the reconstruction algorithm of MAP based on the critical-point filters registration method joined the adaptive regularization parameter so that it can take advantage of thethreealgorithms. After the test,the results show thatcomparedwith the traditional methods, the algorithm of this paper can express the non-translational motion more effectively and the quality of the super-resolution images hasbeen improved obviously.