The convolution layer

A key component of a CNN is the convolution layer, which performs a dot product of a kernel matrix with part of an image and generates an output. This is followed by shifting and repeating the same operation over a complete image and is termed convolution. The region of the input which is taken for the dot product is called the receptive field of the convolution layer. In each convolution layer, there is a set of kernels and they are collectively termed filters.

The input for a convolution layer is an n-dimensional array, meaning the input is an image of the form Width x Height x Depth. For example, if we have a grayscale image of the size 32 x 32, width and height, then the input is 32 x 32 x 1 where depth is the number ...

Get Practical Computer Vision now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.