一种快速的超分辨率图像重构算法

A fast super resolution image reconstruction algorithm

摘要:

基于稀疏表示的超分辨率(SR)图像重构算法需要求 解l₁范数优化问题,时间效率不高。 本文在该方法的基础上加入平移不变性约束条件,并且考虑到稀疏参数对重构结果影响不 大,对重构算法进行简化,避免了求解复杂的l₁范数优化问题,实现了一种快速的图像SR重构。数值实验结果表明,本文算法具有更高的时间效率。

Abstract:

The super resolution (SR) reconstruction of image can be achieved by the locally linear embedding algorithm or the sparse representation algorithm.The locally l inear embedding algorithm requires the number K of the nearest neighbors as a predefined parameter.Improper choice of K will cause the over-fitting or under-fitting of data,which results in blurring of the reconstructed image.The sparse representation algorithm could select the neighbors adaptively,but needs to solve a complicated and time consu ming l₁-norm minimizing problem.In this wo rk,a new super resolution reconstruction algorithm is proprosed by introducing t he shift invariance constraints.Since the sparsity parameter affects the reconst ruction insignificantly,the proposed algorithm is simplified to avoid solving th e minimizing l₁-norm problem,which leads to a signifi cant reduction in the time complexity.The time complexity of the proposed algori thm is O(mn),while that of solving the minimizing l₁-norm by Lasso is O(m³＋nm 2)．Experimental results indicate that the proposed algorithm is m ore efficient than previous algorithms,and holds a good stability in time consum ption with the increasing overlap between contiguous image patches and the dicti onary size.Thus,the proposed algorithm can be implemented with a bigger dictiona ry under acceptable time,which can be adopted for the real time super resolution reconstruction applications.