1 Introductory Material

1.1 New Nodes

The new names are explained relative to the dataset used in the first problem. So read the first problem for more introductory material on these tables

1.2 Convolver

KNIME has a convolution node allowing for two images to be convolved using a number of different algorithms. ### Options

  • Options Window

  • Kernel Column specifies which image is to be used as the kernel (can also be from the kernel creator)
  • The settings for the image are specified here for 2D, select dimensions X and Y, for 3D: X, Y and Z.
  • ‘Out of bounds Selection’ specifies what the algorithm does at the edge of the image
  • BORDER specifies padding with the same values as the edge
  • ZERO VALUE specifies zero padding
  • MIRROR SINGLE will mirror the values at the edge
  • PERIODIC will periodically repeat the values
  • ‘Calculate as Float’ specifies the output as a floating point number (otherwise it might calculate it as the same type as the image which may not be precise enough)

1.2.1 Convolution Settings

  • Settings
  • Here the implementation can be specified, the Fourier-based methods should in most cases be faster (for very small images / kernels they might be slower)

1.3 Convolution Kernel Creator

In the KNIME Image Processing -> IO -> Other, there is a Kernel creator which can be used to specify common kernels as covered in the ‘Image Enhancement’ lecture.

2 Exercises

2.1 Convolution-based Tracking