fitContImage: Data pre-processing and continuous image analysis

Description

This function can be used to analyze a continuous image in .png or .jpeg format, or an image represented as a list object in the format of parnormobs.

Usage

fitContImage(image, gamma.fun = NULL, center = NULL, inimean = NULL, nrun,
 nburn, J, ordering_mu, ordering_sigma, mask = NULL, slice, outputAll)

Arguments

image

This may be a string representing the path to a .png or .jpeg file, or a list object in the same format as par2obs output, with intensity, r.obs, theta.obs, and center the required list contents.

gamma.fun

This is a function, like triangle2 or ellipse, denoting the true boundary. It is optional and only used when the image input refers to a .png or .jpeg file.

center

This is required if the image input refers to a .png or .jpeg file, otherwise it is unused.

inimean

a constant to specify the initial mean functions in the Bayesian estimation.

nrun

the number of MCMC samples to keep for estimation.

nburn

the number of initial MCMC samples to discard.

truncation number of the Gaussian process kernel. The number of eigenfunctions is $2J + 1$.

ordering_mu

Indicates which Gaussian distribution has larger mean intensity: "I", the Gaussian distribution inside the boundary; "O", the Gaussian distribution outside the boundary; "N", no ordering information is available.

ordering_sigma

Indicates which Gaussian distribution has larger intensity variance: "I", the Gaussian distribution inside the boundary; "O", the Gaussian distribution outside the boundary; "N", no ordering information is available.

mask

Logical vector (same length as obs$intensity) to indicate region of interest. Should this data point be included in the analysis?

slice

boolean where TRUE means that slice sampling will be used to sample Fourier basis function coefficients and FALSE means that Metropolis-Hastings will be used instead.

outputAll

boolean controlling the amount of output produced, see value below.

Value

output

If outputAll is FALSE,

estimate

Posterior mean estimate of image boundary at theta values.

theta

A grid of 200 values on $[0,2\pi]$ at which to retrun the estimated boundary.

lower, upper

The lower and upper bounds of a $95\%$ uniform credible band for the image boundary.

If outputAll is TRUE, same as above, and additionally,

musig.smp

posterior samples of $\mu_1$, $\mu_2$, $\sigma_1$, and $\sigma_2$.

coef.smp

posterior samples of Fourier basis function coefficients.

image

the input image passed to fitContImage.

obs

the processed image data passed to BayesBDnormal.

References

Li, M. and Ghosal, S.(2015) "Bayesian Detection of Image Boundaries." arXiv 1508.05847.

Examples

Run this code

# NOT RUN {
set.seed(12345)
gamma.fun = ellipse(a = 0.35, b = 0.25)
norm.obs = parnormobs(m = 100, mu.in = 4, mu.out = 1,
 sd.in = 1.5, sd.out = 1, design = 'J',
 center = c(0.5,0.5), gamma.fun)
norm.samp = fitContImage(image = norm.obs, nrun = 1000, nburn = 0,
 J = 10,ordering_mu = "I",ordering_sigma = "I", slice = FALSE, outputAll = FALSE)
par(mfrow = c(1,3))
plotBD(norm.samp, 1)
plotBD(norm.samp, 2)
plotBD(norm.samp, 3)
# }