Notes 20110210 CIS 6320 Image Processing Talk -- Jamileh Fuzzy Logic
From SnOwy - Ed's Wiki Notebook
- colours can be described as fuzzy sets
- each colours can be given a membership value
- a point in a colour triangle can be given a degree of membership in each item of the colourspace
- fuzzy = unsharp boundaries -- not vagueness (insufficient specificity)
- low level -- preprocessing -- greyness ambiguity
- intermediate-level -- geometric fuzziness
- high-level -- vague knowledge
- improving an image with image fuzzification
- expert knowledge modifies membership of the grey levels of the input image
- the transform domain is the membership plane
- image defuzzification
- fuzzy logic, fuzzy set theory --
- contrast enhancement
- four example approaches
- minimization of fuzziness
- equalization using fuzzy expected value
- compare the grey level with the desired grey level -- use the difference to enhance the image
- fuzzy hyperbolization
- rule-based approach
- no fuzzification used -- used natural language rules
- contrast enhancement -- histogram hyperbolization
- edge detection
- membership function edge detection --
- noise reduction -- realization that noise is a variation of value
- use gradient to decide -- small derivative, most likely noise, big derivative -- edge
- fuzzy derivative values
- image segmentation
- fuzzy clustering
- fuzzy rule-based
- fuzzy thresholding
- fuzzy geometry
- linear vs fuzzy segmentation
- fuzzy segmentation used to create labels for each region
- rule-based example ...
- if dark and neighbour dark and low variation then is this much background.
- membership function moves between pixels to make such a distinction
- fuzzy geometry
- fuzzy area
- fuzzy perimeter
- fuzzy compactness
- fuzzy region labelling
- solves over segmentation
- assign labels with confidence values
- link labels with concepts in ontologies (human)
- best primers in fuzzy image -- Tizhoosh in references