# fields, Tools for spatial data
# Copyright 2004-2011, Institute for Mathematics Applied Geosciences
# University Corporation for Atmospheric Research
# Licensed under the GPL -- www.gpl.org/licenses/gpl.html
"image.smooth" <- function(x, wght = NULL, dx = 1, 
    dy = 1, kernel.function = double.exp, theta = 1, grid = NULL, 
    tol = 1e-08, xwidth = NULL, ywidth = NULL, weights = NULL, 
    ...) {
    # first part of this function is figuring what has been passed and
    # what to do
    if (is.list(x)) {
        # assume that an image list format has been passed as x
        Y <- x$z
        grid <- list(x = x$x, y = x$y)
    }
    else {
        Y <- x
    }
    if (!is.matrix(Y)) {
        stop("Requires a matrix")
    }
    m <- nrow(Y)
    n <- ncol(Y)
    # use information in previous setup kernel function from a
    # a call to setup.image.smooth and in the process override any
    # passed arguments
    if (!is.null(wght)) {
        dx <- wght$dx
        dy <- wght$dy
        xwidth <- wght$xwidth
        ywidth <- wght$ywidth
    }
    # set up grid if it is missing
    if (is.null(grid)) {
        grid <- list(x = (1:m) * dx, y = (1:n) * dy)
    }
    else {
        dx <- grid$x[2] - grid$x[1]
        dy <- grid$y[2] - grid$y[1]
    }
    # padding of zeroes around actual image
    # if less than m and n there may be spurious effects due to
    # the periodicity from the fft.
    # make sure that the span of the kernel is less than xwidth and ywidth.
    # there will be substantial speedup if the kernel has a small support,
    # Y is big  (e.g. 512X512) and Mwidth and N widht are adjusted to suit.
    if (is.null(xwidth)) {
        xwidth <- dx * m
    }
    if (is.null(ywidth)) {
        ywidth <- dy * n
    }
    # kernel wght function as fft
    # reusing this saves an fft for each image smooth.
    if (is.null(wght)) {
        wght <- setup.image.smooth(nrow = m, ncol = n, xwidth = xwidth, 
            ywidth = ywidth, dx = dx, dy = dy, kernel.function = kernel.function, 
            theta = theta)
    }
    M <- nrow(wght$W)
    N <- ncol(wght$W)
    temp <- matrix(0, nrow = M, ncol = N)
    temp2 <- matrix(0, nrow = M, ncol = N)
    # pad with zeroes
    if (!is.null(weights)) {
        temp[1:m, 1:n] <- Y * weights
        temp[is.na(temp)] <- 0
        temp2[1:m, 1:n] <- ifelse(!is.na(Y), weights, 0)
    }
    else {
        temp[1:m, 1:n] <- Y
        temp[is.na(temp)] <- 0
        temp2[1:m, 1:n] <- ifelse(!is.na(Y), 1, 0)
    }
    # temp and temp2 are numerator and denominator of Nadarya-Watson estimator.
    temp <- Re(fft(fft(temp) * wght$W, inverse = TRUE))[1:m, 
        1:n]
    temp2 <- Re(fft(fft(temp2) * wght$W, inverse = TRUE))[1:m, 
        1:n]
    # try not to divide by zero!
    temp <- ifelse((temp2 > tol), (temp/temp2), NA)
    list(x = grid$x, y = grid$y, z = temp)
}