Abstract:
The advancement of machine learning helps for face image processing which is used to
recognize the gender, emotion, and age with deep learning. From the studies of image
recognition for age estimation there is no data set prepared for this task specially the face
image of our society. As we know people intentionally or unintentionally doesn’t tell their
age. For this reason; this study has been performed by using manually collected data set of
our society. For implementing of face image recognition, a large preprocessed data set of
face image should have for training the model. In deep learning; the performance of the
model depends on the amount of data set used for training the model. With limited data set
the model performance becomes degraded. To overcome this problem, we have used
pretrained deep learning model of VGG16 by transfer learning and fine tuning the model.
We trained the fine-tuned model with manually collected data sets of face images by
varying the epoch. The data set has organized in 7 classes grouped in age ranges. In training
the VGG16 fine-tuned model with our data set the model gives good result for prediction
of age. The model performance depends on the size of data set given to the model for
training and the epoch size as observed during training the VGG16 model in this thesis.
From this study the performance of the model had evaluated by using RMSE and R
2
and
the of VGG16 fine-tuned model results; RMSE= 1.74514 and R
2
score=0. 10599.In
addition to this the training loss and validation loss results; loss=3.527 and
val_loss=3.3827.