Neural Networks with R

Step #1 – filtering

The convolutional layer does the heavy math operations. In computer vision, a typical approach to processing an image is to convolute it with a filter to extract only the salient features in it. This is the first operation in a CNN. The input image is applied a filter logic to create an activation map or feature map:

The convoluted feature vector is created by applying the kernel vector on each 3 x 3 vector of the image.

The mathematical steps for filtering are as follows:

1. Line up the feature and the image patch.
2. Multiply each image pixel by the corresponding feature pixel.
3. Add them up.
4. Divide each sum by the total number ...