Projects conducted by our staff were connected with a number of challenges from various fields. Not only image processing, but also pattern recognition and data mining issues had to be tested and combined together. The research resulted in the implementation of general-purpose algorithms in a common library.

Algorithms that were used in object segmentation related tasks:

• color and gray level image segmentation based on cluster analysis, graph theory and pattern recognition techniques (originally developed)
• transformation of image space into a modified HSI space where cluster separation is  preserved
• adaptive threshold determination based on
  • fuzzy sets
  • Otsu method
  • Huang-Wang method and ''fuzzy entropy''
  • Shannon method
• anisotropic-based watershed (originally developed)
• mean shift algorithms for filtering and segmentation
• non-linear low-pass image filters based on
  • adaptive approach and a rotational mask (also called ''smart blur'')
  • anisotropic diffusion
  • image modes enhancements
  • concavity filling operators
  • background substraction
• shape analysis
  • convex hull
  • Ferret and Blair shape coefficient
  • ''Blur shape'' coefficient
• edge detectors (Canny, LoG, Sobel, Kirsch, Robinson, etc.)
• distance transforms
• labeling with run-length objects coding
• image arthmetic library
• and other standard image processing methods

• Hough transformation with original modifications that enhance the detection of round and elliptical objects in noisy images
• contour grouping algorithms  based on the pattern recognition approach
• analysis and filtering of edge maps
• edge tracking
• morphological operators
• feature-based image segmentation

  

More information can be obtained from wojciech.tarnawski@vratis.com  and marcin.smereka@vratis.com