Age estimation based on face images and pre-trained convolutional neural networks

Abstract

Age estimation based on face images plays an important role in a wide range of scenarios, including security and defense applications, border control, human-machine interaction in ambient intelligence applications, and recognition based on soft biometric information. Recent methods based on deep learning have shown promising performance in this field. Most of these methods use deep networks specifically designed and trained to cope with this problem. There are also some studies that focus on applying deep networks pre-trained for face recognition, which perform a fine-tuning to achieve accurate results. Differently, in this paper, we propose a preliminary study on increasing the performance of pre-trained deep networks by applying postprocessing strategies. The main advantage with respect to finetuning strategies consists of the simplicity and low computational cost of the post-processing step. To the best of our knowledge, this paper is the first study on age estimation that proposes the use of post-processing strategies for features extracted using pretrained deep networks. Our method exploits a set of pre-trained Convolutional Neural Networks (CNNs) to extract features from the input face image. The method then performs a feature level fusion, reduces the dimensionality of the feature space, and estimates the age of the individual by using a Feed-Forward Neural Network (FFNN). We evaluated the performance of our method on a public dataset (Adience Benchmark of Unfiltered Faces for Gender and Age Classification) and on a dataset of nonideal samples affected by controlled rotations, which we collected in our laboratory. Our age estimation method obtained better or comparable results with respect to state-of-the-art techniques and achieved satisfactory performance in non-ideal conditions. Results also showed that CNNs trained on general datasets can obtain satisfactory accuracy for different types of validation images, also without applying fine-tuning methods.

Optimization of deep learning features for age-invariant face recognition

“…The model has been trained using more than 2.6 million face images achieving high face recognition rate. Later on, this model has been repeatedly invoked by other research teams, using transfer learning, for age estimation [15,16] and for AIFR [17]. ElKhiyari and Wechsler [17] used this model with an Ensemble of Subspace Discriminant (ESD) classifiers to achieve an accuracy of 80.6% for FGNET dataset and 92.2% for MORPH-II dataset.…”

Estimação de idade em imagens digitais a partir de deep learning para apoiar análise pericial

“…Pode-se resumir que a função de uma camada convolucionalé receber e processar informações tridimensionais produzindo resultados de mesma dimensão [Buduma and Lacascio 2017]. [Anand et al 2017] propôs uma combinação de redes heterogêneas treinadas para extrair recursos de imagens faciais não ideais, ou seja, condições em que ruídos como fundo ou outros objetos estão presentes. Para a estimação de idade, os autores utilizaram o método de regressão aplicadoà uma rede feed forward para gerar um valor numérico real, que por sua vez foi classificado em um grupo de idades.…”

“…Todos estes compostos por imagens de personalidades públicas em situações diversas de iluminação e cenário. [Anand et al 2017] obteve uma acurácia final de 58.49% em um máximo de 2.000épocas, utilizando o Erro Absoluto Médio (MAE) como métrica de validação.…”

“…Para os experimentos, os resultados foram próximos aos trabalhos correlatos apresentados na Seção 3. Como o Experimento 1 buscou dividir as classes em um número próximo aos trabalhos existentes, este projeto superou em 3.51 pontos percentuais o proposto por [Anand et al 2017] e em 1.22 pontos percentuais o proposto por [Rodríguez et al 2017], com 62,00%. Apesar dos Experimentos 2 e 3 apresentarem divisões menores de classes, os resultados alcançados foram considerados como satisfatórios, com valores médios de 73% e 96,85% de precisão.…”

Towards Explainable Face Aging with Generative Adversarial Networks

“…Recently, computational intelligence methods based on Deep Learning (DL) have been increasingly used due to their capabilities of automatically learning data representations [11,12,13,14]. Among DL methods, several approaches to face aging have considered Generative Adversarial Networks (GAN) [15], which have shown high accuracy in automatically generating realistic synthetic images.…”