image.CornerDetectionHarris (version 0.1.2)

image_harris: Find Corners using Harris Corner Detection

Description

An implementation of the Harris Corner Detection algorithm explained at tools:::Rd_expr_doi("10.5201/ipol.2018.229").

Usage

image_harris(
  x,
  k = 0.06,
  sigma_d = 1,
  sigma_i = 2.5,
  threshold = 130,
  gaussian = c("fast Gaussian", "precise Gaussian", "no Gaussian"),
  gradient = c("central differences", "Sobel operator"),
  strategy = c("all corners", "sort all corners", "N corners", "distributed N corners"),
  Nselect = 1L,
  measure = c("Harris", "Shi-Tomasi", "Harmonic Mean"),
  Nscales = 1L,
  precision = c("quadratic approximation", "quartic interpolation", "no subpixel"),
  cells = 10L,
  verbose = FALSE
)

Value

as list of the relevant points with the x/y locations as well as the strenght. Note y values start at the top left corner of the image.

Arguments

x

an object of class magick-image or a greyscale matrix of image pixel values in the 0-255 range where values start at the top left corner.

k

Harris' K parameter. Defaults to 0.06.

sigma_d

Gaussian standard deviation for derivation. Defaults to 1.

sigma_i

Gaussian standard deviation for integration. Defaults to 2.5.

threshold

threshold for eliminating low values. Defaults to 130.

gaussian

smoothing, either one of 'precise Gaussian', 'fast Gaussian' or 'no Gaussian'. Defaults to 'fast Gaussian'.

gradient

calculation of gradient, either one of 'central differences' or 'Sobel operator'. Defaults to 'central differences'.

strategy

strategy for selecting the output corners, either one of 'all corners', 'sort all corners', 'N corners', 'distributed N corners'. Defaults to 'all corners'.

Nselect

number of output corners. Defaults to 1.

measure

either one of 'Harris', 'Shi-Tomasi' or 'Harmonic Mean'. Defaults to 'Harris'.

Nscales

number of scales for filtering out corners. Defaults to 1.

precision

subpixel accuracy, either one of 'no subpixel', 'quadratic approximation', 'quartic interpolation'. Defaults to 'quadratic approximation'

cells

regions for output corners (1x1, 2x2, ..., NxN). Defaults to 10.

verbose

logical, indicating to show the trace of different substeps

Examples

# \dontshow{
mat <- system.file(package = "image.CornerDetectionHarris", "extdata", "building.rds")
mat <- readRDS(mat)
pts <- image_harris(mat)
# }
# \donttest{
library(magick)
path <- system.file(package = "image.CornerDetectionHarris", 
                    "extdata", "building.png")
x    <- image_read(path)
pts  <- image_harris(x)
pts

plt <- image_draw(x)
points(pts$x, pts$y, col = "red", pch = 20)
dev.off()
plt <- image_draw(x)
points(pts$x, pts$y, 
       col = "red", pch = 20, cex = 5 * pts$strength / max(pts$strength))
dev.off()

## Or pass on a greyscale matrix starting at top left
mat <- image_data(x, channels = "gray")
mat <- as.integer(mat, transpose = FALSE)
mat <- drop(mat)
pts <- image_harris(mat)
plt <- image_draw(x)
points(pts$x, pts$y, col = "red", pch = 20)
dev.off()
# }