摘要： 针对现有基于局部线性嵌入的高分辨率图像重建算法对噪声敏感、图像块间边界不连续等问题，提出一种近邻非负线性重建高分辨率图像的流形学习算法。将流形学习重建过程中的近邻线性组合系数约束为非负，并采用基于像素块比例值的特征提取方法。实验结果表明，该算法能重建更多的细节，降低块效应，提高重建图像的峰值信噪比。

关键词: 高分辨率图像重建, 流形学习, 局部线性嵌入, 近邻重建, 图像块, 非负权值

Abstract: The Locally Linear Embedding(LLE)-based high-resolution image reconstruction method has some problems about noise sensitivity and discontinuous boundary between image blocks. Aiming at these problems, this paper proposes a manifold learning method for reconstructing high-resolution image through neighborhood nonnegative linear combination. It restricts linear combination coefficients of neighbors to be nonnegative in the reconstruction process of high resolution images based on manifold learning, and uses a new feature extraction method. Experimental results show that the proposed algorithm can reconstruct more detail features, remove discontinuous boundary and obtain higher Peak Signal Noise Ratio(PSNR).

Key words: high-resolution image reconstruction, manifold learning, Locally Linear Embedding(LLE), neighborhood reconstruction, image block, nonnegative weight value

中图分类号: 

• TN911.73