image augmentations of a matrix, data frame, array or a list of 3-dimensional arrays (where the third dimension is equal to 3)

Augmentation(
  image,
  flip_mode = NULL,
  crop_width = NULL,
  crop_height = NULL,
  resiz_width = 0,
  resiz_height = 0,
  resiz_method = "nearest",
  shift_rows = 0,
  shift_cols = 0,
  rotate_angle = 0,
  rotate_method = "nearest",
  zca_comps = 0,
  zca_epsilon = 0,
  image_thresh = 0,
  padded_value = 0,
  verbose = FALSE
)

Arguments

image: a matrix, data frame, array or list of 3-dimensional arrays where the third dimension is equal to 3
flip_mode: a character string ('horizontal', 'vertical')
crop_width: an integer specifying the new width of the image, after the image is cropped. Corresponds to the image-rows.
crop_height: an integer specifying the new height of the image, after the image is cropped. Corresponds to the image-columns.
resiz_width: an integer specifying the new width of the image, after the image is resized. Corresponds to the image-rows.
resiz_height: an integer specifying the new height of the image, after the image is resized. Corresponds to the image-columns.
resiz_method: a string specifying the interpolation method when resizing an image ('nearest', 'bilinear')
shift_rows: a positive or negative integer specifying the direction that the rows should be shifted
shift_cols: a positive or negative integer specifying the direction that the columns should be shifted
rotate_angle: an integer specifying the rotation angle of the image
rotate_method: a string specifying the interpolation method when rotating an image ('nearest', 'bilinear')
zca_comps: an integer specifying the number of components to keep by zca whitening, when svd is performed
zca_epsilon: a float specifying the regularization parameter by zca whitening
image_thresh: the threshold parameter, by image thresholding, should be between 0 and 1 if the data is normalized or between 0-255 otherwise
padded_value: either a numeric value or a numeric vector of length equal to N of an N-dimensional array. If it's not equal to 0 then the values of the shifted rows or columns will be filled with the user-defined padded_value. Applies only to the shift_rows and shift_cols parameters.
verbose: a boolean (TRUE, FALSE). If TRUE, then the total time of the preprocessing task will be printed.

Value

the output is of the same type with the input (in case of a data frame it returns a matrix)

Details

This function takes advantage of various methods to accomplish image augmentations. The order of the preprocessing steps, in case that all transformations are applied to an image, is : 1st flip image, 2nd crop image, 3rd resize image, 4th shift rows or columns, 5th rotate image, 6th zca-whitening and 7th image-thresholding.

Author

Lampros Mouselimis

Examples


if (FALSE) {

# a matrix
object = matrix(1, 10, 10)

res = Augmentation(object, resiz_width = 8, resiz_height = 8, rotate_angle = 40)


# an array
object = array(0, dim = c(10, 10, 3))

res = Augmentation(object, resiz_width = 8, resiz_height = 8, rotate_angle = 30)


# an array (multiple matrices)
object = array(0, dim = c(10, 10, 10))

res = Augmentation(object, resiz_width = 8, resiz_height = 8, rotate_angle = 20)


# a list of 3-dimensional arrays (where the third dimension is equal to 3)
object = list(array(0, dim = c(10, 10, 3)), array(0, dim = c(10, 10, 3)))

res = Augmentation(object, resiz_width = 8, resiz_height = 8, rotate_angle = 40)
}