摘要：

针对一般单帧图像超分辨率算法运行速度较慢和参数冗余问题,提出一种基于交叉耦合字典,并给出利用随机森林算法解决多元回归问题的单帧超分辨率算法。引出耦合字典中最核心的问题,即从LR图像块到HR图像块的映射问题,得到单帧超分辨率与局部线性回归的关系,设计新的正则化目标函数,并利用随机森林算法优化该目标函数。实验结果表明,与广义相似最近邻域算法、基于实例的邻域回归超分辨率算法等相比,该算法能获得较高的峰值信噪比和结构相似性指标,且所得结果图像纹理更加丰富自然。

关键词: 单帧图像, 超分辨率, 随机森林, 回归矩阵, 正则化

Abstract:

Aiming at the slow running speed and the redundancy of parameters in common single frame Super Resolution(SR) algorithms,a single frame super resolution based on cross coupled dictionary is proposed,in which random forests is used to solve the problem of multi-term regression.The core issue of the coupled dictionary is introduced,that is the mapping problem of LR image patches to HR image patches.The close relationship between single-frame SR and local linear regression is given.A new regularization objective function is proposed,and the algorithm of random forest is used to optimize the regularization objective function.Experimental results show that compared with the algorithm of general Approximate Nearest Neighbour Field(ANNF),Neighbour Regression Example-based SR(NRESR) and the other algorithms,the proposed algorithm can obtain higher value in the aspects of Peak Signal to Noise Ratio(PSNR) and Structural Similarity Index Measurement(SSIM),and the image’s texture is much more natural.

Key words: single frame image, Super Resolution(SR), random forest, regression matrix, regularization

中图分类号: 

• TP391