| Abstract: |  Image registration is an important field of computer
vision. In the last decade, methods using mutual information between
registrered images as their similarity criterion have been gaining
popularity. The aim of this work is to evaluate statistical properties
and speed of certain entropy and mutual information estimator
implementations on data of various dimensionalities and probability
distributions. Among the estimators evaluated are: the histogram
estimator, in its classical form and with enhancements such as
histogram smoothing and adaptive binning, entropy and mutual
information estimator based on kernel density estimation and a
nearest-neighbor based estimator and its faster modifications
replacing nearest-neighbors with approximate nearest-neighbors. Also,
we assess statistical properties of an Renyi entropy estimator based
on the length of a minimum spanning tree spanning the samples.
|
|---|