In this step, we shrink the image stack. For each feature obtained in the convolutional step, we build up a matrix and now find the maximum in each chosen matrix to shrink the entire input. The steps are below:

1. Pick a window size (usually 2 or 3).
2. Pick a stride moving range of pixels (usually 2).
3. Slide the window across the filtered images.
4. For each window, we take the maximum value.

If the slid window does not have the required number of cells as in the previous windows, we take whatever values are available.