In order to build deep neural networks, one modification to the basic convolutional operation that you need to really use is padding.
In this video, I have explained why padding is required and what was the issue with our traditional approach with applying the filter over the original pixel matrix.