RED: RED: Regularization by Denoising

In redR: REgularization by Denoising (RED)

Description

RED: Regularization by Denoising

REgularization by Denoising

Usage

RED(y, x0 = NULL, lambda = 1, sigma = 1, functional = "SR",
  engine = "MF", niter = 50, step = NULL, tol = 0.001, args = NULL)

Arguments

y	cimg object with the observed frame(s)
x0	initial guess for the output image, if NULL an educated guess will be used. If a custom functional is provided this cant be NULL
lambda, sigma	numeric indicating the regularization parameters
functional	character with the optimization task or function with the functional to be used
engine	character indicating the denoised engine or function with the denoiser engine to be used
niter	numeric indicating the maximum number of iterations
step	numeric indicating the step size (if NULL an optimal step size will be used)
tol	numeric indicating the stopping criteria. The algorithm will stop when step < tol. Default = 0.001
args	arguments to be passed implicitly to H HT and f

Examples

im <- lenna
y <- degrade(im, noise = 0.05)
x <- RED(y, sigma = 1, lambda = 5, functional = 'DN', niter = 50)
par(mfrow = c(1,2), mar = c(0,0,2,0)+0.1)
plot(y, interp = FALSE, axes = FALSE, main = 'Degraded im')
mtext(paste(round(PSNR(im, y),2), 'dB'), side = 1, line = -2)
plot(x, interp = FALSE, axes = FALSE, main = 'Restored im')
mtext(paste(round(PSNR(im, x),2), 'dB'), side = 1, line = -2)

## Not run: 
im <- cameraman
y <- degrade(im, blur = 5)
y<- isoblur(im, 3, gaussian = TRUE)
x <- RED(y, sigma = 1, lambda = 4, functional = 'DB', niter = 1500)
par(mfrow = c(1,2), mar = c(0,0,2,0)+0.1)
plot(y, interp = FALSE, axes = FALSE, main = 'Degraded image')
mtext(paste(round(PSNR(im, y),2), 'dB'), side = 1, line = -2)
plot(x, interp = FALSE, axes = FALSE, main = 'Restored image')
mtext(paste(round(PSNR(im, x),2), 'dB'), side = 1, line = -2)

im <- cameraman
L = 2
s <- cbind(c(0,1,2,-2,1,3,-1,-3,-1), c(0,-1,2,1,-2,-3,3,-2,-3))
y <- degrade(im, L = L, s = s, noise = 0.05)
xref <- resize(imsplit(y,'z')[[1]], -100*L, -100*L, interpolation_type = 5)
x <- RED(y, sigma = 1, lambda = 5, functional = 'SR', niter = 50, args = list(scale = L, s=s))
par(mfrow = c(1,2), mar = c(0,0,2,0)+0.1)
plot(xref, interp = FALSE, axes = FALSE, main = 'Bicubic Interpolation')
mtext(paste(round(PSNR(im, xref),2), 'dB'), side = 1, line = -2)
plot(x, interp = FALSE, axes = FALSE, main = 'Super Resolved')
mtext(paste(round(PSNR(im, x),2), 'dB'), side = 1, line = -2)

im0 <- 0.2*pad(cameraman, 256, 'xy')
im1 <- lenna
im2 <- im1 - im0
y1 <- degrade(im1, noise = 0.05)
y2 <- degrade(im2, noise = 0.05)
y0 <- y1 - y2
x0 <- RED(y0, sigma = 1, lambda = 50, functional = 'DN', niter = 100)

par(mfrow = c(1,2), mar = c(0,0,2,0)+0.1)
plot(y0, interp = FALSE, axes = FALSE, main = 'naive')
mtext(paste(round(PSNR(im0, y0),2), 'dB'), side = 1, line = -2)
plot(x0, interp = FALSE, axes = FALSE, main = 'proposed')
mtext(paste(round(PSNR(im0, x0),2), 'dB'), side = 1, line = -2)

## End(Not run)

redR documentation built on May 2, 2019, 3:47 p.m.