What these pictures represent is the output that you get when you apply the two filters from the previous slide (one for detecting horizontal gradients and one for detecting vertical gradients) to the original brick wall image that we started with couple slides back. The result is that we get primitive "edge detection" along both the horizontal and vertical axes that trace out the grout lines from the original image. This is a simple example of how convolutions filters with different weights can be used to identify primitive features in an image – stacking these convolutions together with other non-linearities is the basis of today's deep neural networks image-based tasks.
What these pictures represent is the output that you get when you apply the two filters from the previous slide (one for detecting horizontal gradients and one for detecting vertical gradients) to the original brick wall image that we started with couple slides back. The result is that we get primitive "edge detection" along both the horizontal and vertical axes that trace out the grout lines from the original image. This is a simple example of how convolutions filters with different weights can be used to identify primitive features in an image – stacking these convolutions together with other non-linearities is the basis of today's deep neural networks image-based tasks.