non-parametric denoising and clustering method of noisy images
both indexed by time and space
Description
Two-stage method for the denoising and clustering of stack
of noisy images acquired over time, based on the simple
assumption that a finite sequence of noisy images both indexed
by time and space is composed of noisy versions of only a
limited amount of dynamic features. The aim of the method is
first to denoise signals using both the spatial and temporal
information contained in the data, and then cluster the
denoised signals depending on their dynamic features. Two
signals are considered to have similar features if their
difference does not significantly deviate from zero. By
comparing difference signals, no assumption is therefore made
on the shape of the theoretical signals. In order for the
method to be applicable to experimental data, the data should
be normally distributed (or at least follow a symmetric
distribution) with a constant variance. Also the number of
observations n must be of the form n=d^2. Moreover, the method
is based on the implicit assumption that, for a given data set,
almost each dynamic feature is present in two or more pixels.
The use of the method can be time-consuming depending on the
size of the data-array (see arguments fp.mask.size, and
fp.nproc of the callDenoiseVoxel function in order to reduce
the computation time).