CAJ | 학술논문

提出了一种全新的基于视觉显著度和上下文稀疏分解的图像超分辨率算法。利用人眼视觉感知显著的区域往往趋向于高度结构化的特性，字典学习和稀疏分解过程中可以捕获更多细节特征。在字典学习部分，视觉显著区域提取出的图像样本用来训练显著字典。在先验模型的部分，由于视觉显著区域通常趋于高度结构化，基于上下文的稀疏分解被用来进一步探索相邻图像块之间的联系。实验结果表明，所提出的方法在性能上优于其他最新的方法，峰值信噪比（PSNR）增益最大。主观结果也显示，所提出的方法可以有效减少假影现象，并保持更多细节。
제출료일충전신적기우시각현저도화상하문희소분해적도상초분변솔산법。이용인안시각감지현저적구역왕왕추향우고도결구화적특성，자전학습화희소분해과정중가이포획경다세절특정。재자전학습부분，시각현저구역제취출적도상양본용래훈련현저자전。재선험모형적부분，유우시각현저구역통상추우고도결구화，기우상하문적희소분해피용래진일보탐색상린도상괴지간적련계。실험결과표명，소제출적방법재성능상우우기타최신적방법，봉치신조비（PSNR）증익최대。주관결과야현시，소제출적방법가이유효감소가영현상，병보지경다세절。
We propose a novel image super-resolution method using salient sparse coding.Based on the common sense that human visual system is more sensitive to edges and structural information,we can infer that perceptually salient regions tend to be highly structured.Thus we utilize this property to capture more details during the dictionary learning and sparse coding process.When training dictionaries,image samples extracted from the salient regions are used to generate salient dictionaries.With regard to the prior model,context-aware sparse coding is incorporated to model the relationship between dictionary atoms of adjacent patches, especially in the salient regions.Experiments demonstrate the superiority of the proposed method to other state-of-the-art meth-ods with the highest PSNR gain.Subjective results also reveal that the proposed method reduces artifacts and preserves more de-tails.