【摘要】 随着模式识别和计算机视觉的发展,根据人脸图像自动进行年龄估计在人机交互、安全监控和娱乐等领域已经成为一个非常热门的话题。针对特征冗余及对所提取特征不能充分利用的问题,构建一种基于单模型集成的神经网络框架(Age Estimation Framework Based on Single-Model Integration,AEF-SMI)。首先使用5×5、3×3和2×2的级联卷积核提取丰富的空间结构信息,连接不同卷积核组成不同通道,通过集成不同通道获取不同深度卷积激活特征,使网络框架获取高层的语义信息的同时也获取低层边缘纹理信息,最后利用提取到的特征对图片进行年龄估计。实验结果表明,在IMDB-WIKI与Group数据库上,与主流的年龄分类算法相比,AEF-SMI框架得到的准确率更高。更多还原

【Abstract】 With the development of pattern recognition and computer vision, automatic age estimation based on face images has become a hot topic in the fields of human-computer interaction, security monitoring and entertainment. Aiming at solving the problem of feature redundancy and the inability to fully utilize the extracted features, an age estimation framework based on single-model integration(AEF-SMI) is constructed. Firstly, 5×5, 3×3 and 2×2 cascading convolution kernels are used to extract rich spatial structure information, connect different convolution kernels to form different channels, and integrate different channels to obtain different depth convolution activation features, so that the network framework obtaining the high-level semantic information also acquires the low-level edge texture information. Finally, use the extracted features to estimate the age of the image. The experimental results show that the AEF-SMI framework has higher accuracy than the state-of-art classification algorithms on the IMDB-WIKI and Group databases.更多还原