Abstract: Example regression-based technique has been recognized as a simple but effective image Super-Resolution （SR） method. However, simple linear regression model often cannot represent the complex relationship between Low-Resolution（LR） and High-Resolution（HR） images well. At the same time, it can pose a memory issue that when the dictionary and regression size can grow to more than a gigabyte, limiting applicability in memory constrained scenarios. To address those problems, it proposes an optimized example-based SR method with weighted-feature via AdaBoost. In the training stage, it begins with the learning of a sparse dictionary as anchored points from a training set using [K]-SVD dictionary learning algorithm. And then the anchored neighborhood regression model is employed to build a set of strong regressors by using the AdaBoost regression algorithm with [T] rounds. Finally, the learned regressors are coded as a linear combination of few basis regressors for SR reconstruction. To verify the effectiveness of the proposed SR method, four publicly available datasets are used to compare the SR performance with other state-of-art methods. The experimental results show that the proposed algorithm gains better reconstruction performance and lower memory usage, as well as it’s better to the compared algorithms in terms of both objective and visual quality assessments.

Key words: AdaBoost, anchored neighbor regression, dictionary learning, [K]-SVD

摘要： 实例回归是一种简单但有效的超分辨重建方法。然而，简单线性回归器不仅不能很好地表征低分辨与高分辨图像之间复杂的非线性关系，同时，在字典和回归器规模较大的情况下，存在内存占用过高的现象，限制了该类方法在内存受限情况下的适用性。针对这些问题，提出了一种自适应特征增强的实例回归超分辨率重建优化方法。该方法利用[K]-SVD字典学习算法从训练集中学习一个稀疏字典作为锚点；利用锚点邻域回归通过[T]次自适应增强算法得到一组强回归器；将得到强回归器进行优化编码，得到一组回归基和其相应的编码系数用于超分辨重建。为验证提出算法的有效性，分别与其他主流方法在四个公共标准数据集上进行超分辨对比实验。实验结果表明，提出的方法在客观质量和视觉质量评价两个方面上均取得了较好的重建质量，具有较好的重建性能和较低的内存占用。

关键词: AdaBoost, 锚点邻域回归, 字典学习, [K]-SVD