基于双注意力机制和元迁移学习的个性化图像美学评价方法

Personalized Image Aesthetic Evaluation Method Based on Dual Attention Mechanism and Meta-Transfer Learning

个性化图像美学评价针对不同用户之间的个性化审美差异进行感知评估,取得了广泛的应用。然而,目前存在的大众化图像美学模型无法很好地适应小样本个性化图像美学评价任务。为解决该问题,提出了一种融合双注意力机制的EfficientNet网络和元学习的PIAA方法(DA-EBLG-PIAA),将单个用户的个性化打分分别组成不同的单个任务,使用EfficientNet网络作为主干网络,适应小样本学习任务,并融合了双注意力机制,更好地捕捉了空间和通道维度中的全局特征依赖关系。实验结果表明提出的个性化美学评价方法性能优于许多当前存在的模型,可以有效地应用于个性化图像美学感知评价。

Personalized image aesthetic evaluation has been widely used for perception evaluation of different users'personalized aesthetic differences.However,the existing popular image aesthetic model can not adapt well to the small sample personalized image aesthetic evaluation task.To solve this problem,this paper proposes the dual attention EfficientNet network and meta-learning PIAA method(DA-EBLG-PIAA),which combine the personalized ratings of a single user into different single tasks respectively.The EfficientNet network serves as the backbone network to accommodate small sample learning tasks.The dual attention mechanism is integrated to better capture the global feature dependencies in space and channel dimensions.The experimental results show that the performance of the proposed personalized aesthetic evaluation method is better than many existing models and can be effectively applied to the aesthetic perception evaluation of personalized images.