Filter / Kernel
/ˈfɪltər ˈkɜːrnəl/
A small matrix of learnable weights that slides over input during convolution to detect specific features.
In italiano: Filtro / KernelFilters are learned during training to detect useful features. Common sizes: 3x3, 5x5, 7x7. Multiple filters extract different features. The term 'kernel' is used interchangeably with filter.
Examples
- 3x3 edge detection filter
- 5x5 Gaussian blur kernel
- Learned feature detectors