---
type: learning
area: learning
status: learning
date: 2026-05-03
created: 2026-05-03
updated: 1980-01-01
tags:
  - learning
---
- Image Filter = Image kernel
Small matrix applied to an image
Applying a kernel to an image means :
- Example
This is an image
![[img-20240930-205733-15d68653.png]]
- This a filter
![[img-20240930-205753-fd9b72fc.png]]
- When we have filter on top of image we multiple those values
![[img-20240930-205909-e20d7397.png]]
- we can edit stride distance means how much pixel we move the filter on the top of image 
- Kernels are not always 3x3
- When doing convolution images we lose borders in images but if we want to preserve it we do padding like adding pixel with 0 value all around image
### Why not using ANNs instead of CNN?
![[img-20240930-212954-bdbd7f42.png]]
### Convolutional Layer 
focuses of on local connection neurons are only  connected to one next neuron

- Pooling layers used to reduce number of parameters resulted from conv layer
- ![[img-20240930-215249-d746d098.png]]
- Max Pooling
- ![[img-20240930-220308-99005450.png]]
- In it we grab the max value in the window, even a small pooling "kernel" of 2 by 2 with stride of 2 will remove 75% of input data.
	There is also average pooling

- Dropout
can be thought as form of regularization to help preventing over-fitting mean when training units are dropped , along with their connections