面向印刷质量控制的元学习盲图像质量评价方法

Meta-Learning Based Blind Image Quality Assessment for Printing Quality Controlling

中文关键词: 图像质量评价  元学习  图像处理  印刷质量控制#$TAB

英文关键词: image quality assessment  meta-learning  image processing  printing quality control

中文摘要:

      目的 为了更好地检测印刷图文复制效果，提高生产效率，提出一种针对图像复杂失真和内容变化的元学习盲图像质量评价模型。方法 首先在元训练部分，通过ResNet50网络获取多个失真数据集的共有失真先验知识;然后在元测试部分，融合ResNet50的多层次特征，实现对图像局部失真和全局失真的完整描述;最后通过特征降维、融合获得多层次特征的权值，建立图像质量评价网络模型。结果 模型在真实失真数据集LIVEC上SRCC达到0.87以及在合成失真数据集LIVE上SRCC达到0.97，且模型的预测性能和泛化性能都要优于其他算法。结论 所提出的元学习盲图像评价方法能够准确预测不同类型图像质量分数，可为印刷图像质量评价和印刷生产控制提供一定指导。

英文摘要:

      To better detect the effect of printed images and improve the production efficiency, a meta-learning based blind image quality assessment model for complex distortion and content changes of images is proposed in this paper. First, in the meta-training part, the shared distortion prior knowledge of multiple distortion data sets is obtained through the ResNet50 network; then, in the meta-testing part, the multi-level features of ResNet50 are merged to achieve a complete description of the image local and global distortion; finally, through the feature dimension reduction and fusion to obtain the multi-level feature weights, a network model for image quality assessment is established. A large number of experimental results have proved that the prediction performance and generalization performance of the proposed model are better than other algorithms on the authentic distortion dataset (LIVEC) with SRCC of 0.87 and the synthetic distortion dataset (LIVE) with SRCC of 0.97. The proposed meta-learning based blind image quality assessment method can accurately predict the quality scores of different types of images, which can provide certain guidance for the quality evaluation of printing images and the control of printing production.