Module imagingbook_plugins_book
Package Ch08_Binary_Regions

Class Flusser_Moments_Matching_Demo

java.lang.Object
Ch08_Binary_Regions.Flusser_Moments_Matching_Demo
All Implemented Interfaces:
PlugIn, JavaDocHelp
public class Flusser_Moments_Matching_Demo extends Object implements PlugIn, JavaDocHelp

This ImageJ plugin demonstrates the use of complex invariant Flusser moments for 2D shape matching. It opens a special binary input image (collage of shapes), where the top row are the reference shapes. They are assigned different colors. All other shapes are compared to the reference shapes and colored with the color of the most similar reference shape. See Sec. 8.6.5 of [1] for additional details.

The plugin offers two options for finding the "closest" reference moment vector for a given shape:
(a) Mahalonobis distance (MD, see Eq. 8.57 and Sec. G.3 of [1]): The MD is calculated from the (pre-calculated) covariance matrix of the moment vectors of the entire 'Kimia1070' dataset to obtain representative statistical shape parameters. The MD compensates the large magnitude variation between the 11 moment space dimensions.
(b) Euclidean distance (ED): Here the distance between moment vectors is measured using the ED. The large-magnitude moment elements obviously dominate this quantity. Typically only the 3-4 largest moments are actually used, all others have no effect. Thus results with the ED are inferior to the ones obtained with the MD.
In both cases matching is performed linearly, i.e., a given moment vector is simply compared to all reference vectors.

[1] W. Burger, M.J. Burge, Digital Image Processing – An Algorithmic Introduction, 3rd ed, Springer (2022).

Version:
2023/01/02
See Also: