针对许多评价方法都需要大量有效的训练样本离线学习得到特征提取器的问题, 同时考虑到人脑能自然感知数据的内在低维特征, 这种感知恰以流形方式存在, 提出了一种采用在线流形学习的彩色图像质量评价方法。实验结果表明, 所提出方法在LIVE、CSIQ和TID2008三个数据库的平均Spearman秩相关系数(SROCC)达到0.91。相比于其它方法, 在线流形学习的彩色图像质量评价方法与主观视觉感受吻合度更高。

Aiming at feature extractors for the majority of IQA obtained by offline learning from a large number of effective training samples, taking into account the fact, the brain can perceive inherent low dimensional characteristics, which exists in terms of manifold. Therefore, a novel color IQA method based on online manifold learning is proposed. The experimental results show that the proposed method has good performance in the LIVE, CSIQ and TID2008 image database, and the average of Spearman Rank Order Correlation Coefficient (SROCC) is more than 0.91. Compared with some recent state-of-the-art IQAs, the assessment obtained by the proposed method is consistent with the subjective perception.