Convolution
Convolution involves a few concepts, such as convolve, stride, and padding.
For two-dimensional images, convolve happens for each of the color channels. Suppose you have a weight matrix and the image (shown as values at each pixel) as shown in the following figure.
The weight matrix (often called kernel or filter) is applied to the image by placing the kernel over the image to be convolved and moving it across the entire image. If the weight matrix moves 1 pixel at a time, it is called a stride of 1. At each placement, the numbers (pixel values) from the original image are multiplied by the number of the weight matrix that is currently aligned preceding to it.
The sum of all these products is divided by the kernel's normalizer. ...
Become an O’Reilly member and get unlimited access to this title plus top books and audiobooks from O’Reilly and nearly 200 top publishers, thousands of courses curated by job role, 150+ live events each month,
and much more.
Read now
Unlock full access