摘要
针对多帧低分辨率图像重建问题,提出了基于稀疏编码和随机森林的超分辨率算法。首先,使用高分辨率训练图像和低分辨率训练图像获取高分辨率字典;然后,使用重叠块缓解块边界的振铃现象,并使用反向投影保证全局一致性;最后,利用稀疏编码提取和融合LR图像中的有用信息,随机森林完成分类。实验结果表明,相比其他几种较新的超分辨率算法,本文算法重建获得的峰值信噪比(PSNR)最高,重建后的图像最为自然,且具有较快的运行速度。
For the problem of multiple low resolution images reconstruction, a multi-frame image super resolution algorithm using sparse coding and random forest is proposed. Firstly, high resolution training images and low-resolution images are used to obtain high resolution in the dictionary. Then, the ringing of the block boundary is alleviated by overlapping blocks, and the whole consistency is guaranteed by reverse projection. Finally, sparse coding is used to extract useful information from LR images, and Random forest is used to finish classification. Experimental results show that proposed algorithm has the highest PSNR comparing with several other advanced algorithms. It has the most natural reconstruction image and the fastest execution time.
出处
《电子设计工程》
2017年第8期158-162,共5页
Electronic Design Engineering
关键词
图像重建
超分辨率
稀疏编码
随机森林
多帧
反向投影保
image resolution
super-resolution
sparse coding
random forest
muhi-frame
reverse projection
