Application of Linear Scale Space and the Spatial Color Model in Microscopy

Structural features and color are used in human vision to distinguish features in light micorscopy. Taking these structural features and color into consideration in machine vision often enables a more robust segmentation than based on intensity tresholding. Linear scale space theory and the spatial color model provide a powerful framework for feature extraction in microscopy images. Differential geometry is applied in image analysis by convolving the image with Guassian derivatives of the appropriate scale for the objects of interest.

Grayscale microscopy images acquired with a B/W camera contain structrual features for which grayscale scale space provides a robust detection framework. The work of Jan Koenderink provides the theoretical basis for this approach. From this work we know that the linear scale space framework offers robust structrural feature selection for image distortion, noise, and intensity changes. Feature detectors can be constructed based on differential invariants, which are relatively insensitive to changes in illumination condition and signal to noise ratio, which is an important feature in light microscopy. Several differential invariants in light microscopy can be used.

For color light microscopy we use the spatial color model as proposed by Jan Koendering and Jan-Mark Geusebroek to select different colored regions and objects. Differential invariants can be constructed which are insensitive to changes in illumination, color temperature and illumination intensity.

Scalespace and the spatial color model provide the biology microscopist with a powerful and intuititive framework for the detection of features in microscopy images.