Why the validation error is lower than training error

[VikrantKarale]

I have modified a original code so that I can plot training vs validation loss curves for the losses obtained after each epoch. But, I am getting validation loss less than training loss. I am plotting average training and validation loss (total loss divided by total images). The plot is shown below: Can anybody tell why it might be happening?

[SerajRzmi]

in my opinion this is not overfitting but I have some recommends if you use NN: try and test these:

1. don't use drop out in primary layers
2. use low drop out or regularization(L1,L2) rate