Noise removal, image of a face

KEA is a flexible toolbox of robust methods for image, data-processing and for computational geometry tasks. It has been developed by academics and applied researchers of Swansea University and the University of Nottingham in the UK and of the National University of Tucumán in Argentina.

Multiscale Medial Axis

  • Convex Hulls and Directional Convex Hulls
  • Convex Envelopes and Directional Convex Envelopes
  • Tight Smoothing of Singular Functions
  • 2D Multiscale Medial Axis
  • 3D Multiscale Medial Axis Surface Reconstruction
  • Scattered Data Approximation

Shape Interrogation

Our methods are fast and do not require any information on mesh connectivity, topology, parameterisation or differentiability properties. and are stable under sparse sampling measured by Hausdorff distance.

Applications include Detection of High Curvature Points and Junction Points,  Detection of Termination Points, Detection of Curve to Curve Intersections, Detection of Curve to Surface Intersections, Detection of Curve to Solid Intersections, Detection of Curves to Surfaces to Solid Intersections, Detection of Surface to Surface Intersection, Detection of Surface to Solid Intersection, Detection of Surface to Surface to Solid Intersection, Detection of Solid to Solid Intersection, Detection of Small Geometrical Gaps Between Geometric Elements, Detection of Features in Point-Clouds.

Signal Processing

Many of our higher dimensional methods apply to signal processing. We show a special application to high oscillation area extraction and spike detection.

Image Processing

(Anisotropic) Expansion and Contraction, Removal of Thin Lines and Scratches, Impulse Noise Removal, Image Interpolation (Inpainting): Old Image, Restoration & Text Removal, Thin objects enhancement, One-sided Image Smoothing and Geometric, Watermarking, Stable and Affine, Invariant Detection of, Ridges and Valleys, Stable and Affine Invariant Edge Detection, Corner Detection, Neck, Blobs and Small Objects, Detection High Oscillation and Rough Area Detection.

These methods are based on novel tools that depend exclusively on the geometric structure of the image and on the particular task to be performed on the image. As such, they are fundamentally different from those methods that require comparing pixel values in a predetermined mask using some ad hoc problem-designed convolution function, or that require the solution of ad hoc problem-dependent partial differential equations.