- For calculating the features various filters with various filter sizes and parameters are applied to the region around this pixel (2D Kernels).
- Results are concatenated and yield the final feature vector describing the pixel.
Intellesis Basic Features
Basic Features 25
Used Filters:
- Gaussian filter (5 different sigma) = 5 feature dimensions
- Sobel filter (5 sigma) = 5 feature dimension
- Gabor filter (6 theta, 1 different sigma, 2 different frequencies) = 12 feature dimensions
- Hessian filter (1 sigma) = 3 feature dimensions (one for derivative in direction xx, one for derivative in direction xy and one for derivative in direction yy)
Basic Features 33
Used Filters:
- Gaussian filter (20 different sigma) = 20 feature dimensions
- Sobel filter (1 sigma) = 1 feature dimension
- Gabor filter (1 theta, 2 different sigma, 2 different frequencies) = 4 feature dimensions
- Mean filter (5 different sizes) = 5 feature dimensions
- Hessian filter (1 sigma) = 3 feature dimensions (one for derivative in direction xx, one for derivative in direction xy and one for derivative in direction yy)