Here is an example of edge detection using a kernel (filter) or feature detector in terms of machine learning context. You can play with the kernel rotate it, change the values in the cell and see how the image is getting affected.
Check out the Source Code also written by me, it’s open source.
Learn about Prewitt operator and Sobel operator to customize the above kernel with known heuristics.
This technique is useful since instead of feeding the whole image as an array in a neural network we can detect edges and other features by applying various types of techniques and feed that instead.
Info
I will continue to write here as I learn about CNN more in my free time.