In simple words: A CNN learns reusable visual patterns from nearby pixels.
Local scanning
A Convolutional Neural Network (CNN) uses small learnable filters that scan an image. The same filter is applied across many positions, helping it find a feature wherever it appears.
Layers build features
An early layer may respond to edges. Later layers combine edges into shapes and parts. A final classifier can use those features to predict an image category. CNNs can also support detection and other vision tasks.
See the flow
image→convolution filters→feature maps→prediction
Simple example
A filter that responds to vertical edges may find the side of a door in different positions of a photo.
Remember: A CNN learns reusable visual patterns from nearby pixels.