Abstract: Facial age estimation is an important task in the field of computer vision and is of great significance for applications such as face recognition and crowd analysis. Accurately extracting features from facial pictures is a critical and challenging task in age assessment. Traditional evaluation methods use ordered regression based on CNN, but when predicting adjacent ages, the method of convolutional neural network is difficult to capture the global feature representation, resulting in a decrease in prediction accuracy. In order to solve this problem, this paper proposes a new age assessment method that combines the improved Cloformer model with ordered regression. Compared with traditional ordered regression based on convolutional neural networks, Cloformer uses a self-attention mechanism to better capture the relationship between different regions in the image when capturing image features, so as to better learn the feature differences between adjacent ages. The method proposed in this article first optimizes the Cloformer model, and then combines the optimized Cloformer model with ordered regression to better utilize age sequence information and achieve more accurate age prediction, and then trains through end-to-end optimization The improved Cloformer model and ordered regression model enable the proposed method to better learn the relationship between facial features and age sequences. Finally, the superiority of the proposed method is verified by experiments on multiple public datasets.

Key words: facial age estimation, computer vision, feature extraction, ordinal regression, facial features

摘要： 年龄评估是计算机视觉领域中的一个重要任务，对于人脸识别、人群分析等应用具有重要意义。在年龄评估中，准确提取面部图片特征是关键且具有挑战性的任务。传统评估方法采用基于卷积神经网络的有序回归，然而在预测相邻年龄时，卷积神经网络难以捕获全局特征表示，导致预测精度下降。为了解决该问题，本文提出一种新的年龄评估方法，将改进的Cloformer模型与有序回归相结合。相较于传统的基于卷积神经网络的有序回归，Cloformer在捕捉图像特征时能够利用自注意机制更好地捕捉图像中不同区域之间的关系，从而更好地学习相邻年龄之间的特征差异。本文所提出的方法首先对Cloformer模型进行了优化，然后将优化后的Cloformer模型与有序回归相结合，以便更好地利用年龄序列信息，实现更精准的年龄预测，接着通过端到端优化训练改进后的Cloformer模型和有序回归模型，使所提方法更好地学习面部特征和年龄序列关系。最后，通过在多个公开数据集上进行实验，验证了所提方法的优越性。

关键词: 年龄评估, 计算机视觉, 特征提取, 有序回归, 面部特征