In computer vision, estimating face age from face images is an important but challenging task. In this paper, we put forward a general framework called deep multilevel feature fusion network. This deep convolution neural network framework can combine the features extracted from convolution layers of different depths, which can not only ensure the heterogeneity of features, but also ensure the complementarity of features, and can estimate age attributes from the face image. We finetune the deep convolution neural network to a deep level feature fusion network, and use the age classifier to classify the age-related features of the face extracted by the new feature extraction method. This strategy can incorporate the information extracted from face image more effectively, and further improve the ability of age estimation. The highly discriminative and compact classification framework enables the classification accuracy to reach the advanced level on several face image datasets.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Deep Multi-level Feature Fusion Network for Age Estimation


    Contributors:
    Chen, Liming (author) / Jiang, Qiang (author) / Huang, Chuan (author)


    Publication date :

    2021-10-20


    Size :

    952569 byte




    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Feature-level sensor fusion

    Peli, T. Young, M. Knox, R. Ellis, K. K. | British Library Conference Proceedings | 1999


    Recursive building of local maps using multi level feature fusion

    Richter, Eric / Scheunert, U. / Wanielik, G. | Tema Archive | 2007


    Intelligent non-pneumatic tire health monitoring method based on multi-feature fusion deep neural network

    DENG YAOJI / LU KEYU / WANG ZHIYUE et al. | European Patent Office | 2024

    Free access


    Feature-level fusion in personal identification

    Yongsheng Gao, / Maggs, M. | IEEE | 2005