Smooth a 2d field based an a neighbourhood radius

nbhd_smooth takes a 2d field, and for each pixel in that field a square neighbourhood is constructed with a specified radius - that is the number of pixels to the left/right and above/below the pixel at the neighbourhood centre. For each pixel, the mean value of the pixels in the neighbourhood is returned resulting in a smoothed field on the same grid as the original field.

Usage

nbhd_smooth(
  x,
  radius,
  threshold = NA,
  comparator = c("ge", "gt", "le", "lt", "between", "outside"),
  include_low = TRUE,
  include_high = TRUE,
  boundary = c("zero_pad", "missing")
)

cumsum_2d(
  x,
  threshold = NA,
  comparator = c("ge", "gt", "le", "lt", "between", "outside"),
  include_low = TRUE,
  include_high = TRUE
)

nbhd_smooth_cumsum(x, radius, boundary = c("zero_pad", "missing"))

Arguments

x: A 2d array, geofield or geolist.
radius: The radius of the neighbourhood in pixels.
threshold: A threshold for computing binary probabilities. If comparator = "between", it must be a two element vector. Set to NA to use the raw data.
comparator: How to compare x with the threshold to compute binary probabilities. Can be "ge", "gt", "le", or "lt" for >=, >, <= and < respectively. Can also be "between" or "outside", in which case the binary probability of being between or outside of the two values given in threshold is computed.
include_low: Logical. Whether to include to the lower of the two thresholds in the comparison when comparator = "between" or comparator = "outside".
include_high: Logical. Whether to include to the higher of the two thresholds in the comparison when comparator = "between" or comparator = "outside".
boundary: How to treat the boundaries. Current options are "zero_pad" and "missing". See Details for more information on these options.

Value

A smoothed 2d field. In the case of geofields and geolists, the attributes are retained.

Details

For pixels that are closer to the edge of the input field than the radius there are two options. The default is to zero pad the field outwards with pixels outside of the original field set to zero. If boundary = "missing" the pixels closer to the edge than the radius will be set to missing.

The neighbourhood means are computed by first calculating the cumulative sum of the array from the top left to the bottom right and then computing the total for each neighbourhood by subtraction. This means that the algorithm is equally efficient regardless of the neighbourhood size. It is possible to compute each stage at a time with fist cumsum_2d and then nbhd_smooth_cumsum.

If a threshold is given, the binary probability of the input field compared with the threshold is computed at each pixel before applying the function. By default the comparison is x >= threshold, but other comparisons can be chosen using the comparator argument.

Examples

# Create a 2d array
z <- array(dim = c(100, 100))
for (i in 1:100) {
  for (j in 1:100) {
    z[i, j] <- sin(i / 10) + sin(j / 10)
 }
}

image(z, zlim = range(z), col = heat.colors(255))

# With zero-padding the input array
image(nbhd_smooth(z, 5), zlim = range(z), col = heat.colors(255))

image(nbhd_smooth(z, 10), zlim = range(z), col = heat.colors(255))

image(nbhd_smooth(z, 20), zlim = range(z), col = heat.colors(255))


# Without zero padding
image(nbhd_smooth(z, 5, boundary = "missing"), zlim = range(z), col = heat.colors(255))

image(nbhd_smooth(z, 10, boundary = "missing"), zlim = range(z), col = heat.colors(255))


# Add a threshold
image(nbhd_smooth(z, 10, 0.5), zlim = c(0, 1), col = heat.colors(255))

image(nbhd_smooth(z, 10, 0.5, "lt"), zlim = c(0, 1), col = heat.colors(255))

image(nbhd_smooth(z, 10, c(-0.5, 0.5), "between"), zlim = c(0, 1), col = heat.colors(255))

image(nbhd_smooth(z, 10, c(-0.5, 0.5), "outside"), zlim = c(0, 1), col = heat.colors(255))