Convolutional Neuronal Networks (CNN) are extensively employed in image classification. The following project acts as a basic classifier of Hand-Written characters ranging from 0 to 9 and from A to Z (including lowercase and capital letters). I used three different types of networks in this model:
- A dense one (Dense)
- One without DropOut (CNN)
- One with DropOut (CNN&DO)
Here I present the script used for the CNN&DO model
Where one can see that the input images dimensions are 28x28 px in black and white scale. We apply a set of 32 and 64 convolutional filters in addition to a 2x2 MaxConvolution matrix, a 50% DropOut is used to avoid overfitting the model.
Below is presented a comparison between the three models and its respectively accuracy values
It's easy to see that the CNN&DO has a better performance than the CNN does
This shows that CNN are better than conventional Dense Neural Networks for picture classification. Furthermore, the continuation of this project could involve word classification alternatively.