【摘要】 针对现有算法中字典训练花费的时间非常大,且超分辨率图像重建效果不够理想等问题,提出一种改进型字典学习的超分辨率图像重建算法.该算法在字典训练阶段,先采用PCA对低频样本集进行降维,再单独训练出低频字典,然后利用稀疏表示系数集和高频图像样本集训练出对应的高频字典,从而提高了字典构建速度.在重建阶段,先利用字典重建出初始的高分辨率图像,再根据图像结构自相似的特征,对图像进行结构自相似优化,然后对图像进行全局优化,从而提高了重建图像的质量.实验结果表明,该方法无论是在客观评价指标还是主观视觉效果方面都有明显的提高.更多还原

【Abstract】 In order to overcome the deficiency of existing algorithm,such as long training time and unsatisfactory reconstruction result,an improved image super-resolution reconstruction algorithm is proposed in the paper. In this algorithm,low frequency dictionaries are trained after reducing the dimensionalities of low frequency samples by PCA. Then,high frequency dictionaries are trained with sparse coefficients and samples of high frequency images. In the stage of reconstruction,the initial high frequency image is first structured by using dictionaries. Then optimization results are calculated based on the structure of selfsimilarity. Finally,high quality images are obtained after global optimization. The results show that the method not only has a high score in objective evaluation index,but also has a perfect vision effect.更多还原