#######################################################

# Copyright (c) 2015, ArrayFire

# All rights reserved.

#

# This file is distributed under 3-clause BSD license.

# The complete license agreement can be obtained at:

# http://arrayfire.com/licenses/BSD-3-Clause

########################################################

"""

Image processing functions for arrayfire.

"""

from .library import *

from .array import *

from .data import constant

import os

def gradient(image):

"""

Find the horizontal and vertical gradients.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

Returns

---------

(dx, dy) : Tuple of af.Array.

- `dx` containing the horizontal gradients of `image`.

- `dy` containing the vertical gradients of `image`.

"""

dx = Array()

dy = Array()

safe_call(backend.get().af_gradient(ct.pointer(dx.arr), ct.pointer(dy.arr), image.arr))

return dx, dy

def load_image(file_name, is_color=False):

"""

Load an image on the disk as an array.

Parameters

----------

file_name: str

- Full path of the file name on disk.

is_color : optional: bool. default: False.

- Specifies if the image is loaded as 1 channel (if False) or 3 channel image (if True).

Returns

-------

image - af.Array

A 2 dimensional (1 channel) or 3 dimensional (3 channel) array containing the image.

"""

assert(os.path.isfile(file_name))

image = Array()

safe_call(backend.get().af_load_image(ct.pointer(image.arr),

ct.c_char_p(file_name.encode('ascii')), is_color))

return image

def save_image(image, file_name):

"""

Save an array as an image on the disk.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image.

file_name: str

- Full path of the file name on the disk.

"""

assert(isinstance(file_name, str))

safe_call(backend.get().af_save_image(ct.c_char_p(file_name.encode('ascii')), image.arr))

return image

def load_image_native(file_name):

"""

Load an image on the disk as an array in native format.

Parameters

----------

file_name: str

- Full path of the file name on disk.

Returns

-------

image - af.Array

A 2 dimensional (1 channel) or 3 dimensional (3 or 4 channel) array containing the image.

"""

assert(os.path.isfile(file_name))

image = Array()

safe_call(backend.get().af_load_image_native(ct.pointer(image.arr),

ct.c_char_p(file_name.encode('ascii'))))

return image

def save_image_native(image, file_name):

"""

Save an array as an image on the disk in native format.

Parameters

----------

image : af.Array

- A 2 or 3 dimensional arrayfire array representing an image.

file_name: str

- Full path of the file name on the disk.

"""

assert(isinstance(file_name, str))

safe_call(backend.get().af_save_image_native(ct.c_char_p(file_name.encode('ascii')), image.arr))

return image

def resize(image, scale=None, odim0=None, odim1=None, method=INTERP.NEAREST):

"""

Resize an image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

scale : optional: scalar. default: None.

- Scale factor for the image resizing.

odim0 : optional: int. default: None.

- Size of the first dimension of the output.

odim1 : optional: int. default: None.

- Size of the second dimension of the output.

method : optional: af.INTERP. default: af.INTERP.NEAREST.

- Interpolation method used for resizing.

Returns

---------

out : af.Array

- Output image after resizing.

Note

-----

- If `scale` is None, `odim0` and `odim1` need to be specified.

- If `scale` is not None, `odim0` and `odim1` are ignored.

"""

if (scale is None):

assert(odim0 is not None)

assert(odim1 is not None)

else:

idims = image.dims()

odim0 = int(scale * idims[0])

odim1 = int(scale * idims[1])

output = Array()

safe_call(backend.get().af_resize(ct.pointer(output.arr),

image.arr, ct.c_longlong(odim0),

ct.c_longlong(odim1), method.value))

return output

def transform(image, trans_mat, odim0 = 0, odim1 = 0, method=INTERP.NEAREST, is_inverse=True):

"""

Transform an image using a transformation matrix.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

trans_mat : af.Array

- A 2 D floating point arrayfire array of size [3, 2].

odim0 : optional: int. default: 0.

- Size of the first dimension of the output.

odim1 : optional: int. default: 0.

- Size of the second dimension of the output.

method : optional: af.INTERP. default: af.INTERP.NEAREST.

- Interpolation method used for transformation.

is_inverse : optional: bool. default: True.

- Specifies if the inverse transform is applied.

Returns

---------

out : af.Array

- Output image after transformation.

Note

-----

- If `odim0` and `odim` are 0, the output dimensions are automatically calculated by the function.

"""

output = Array()

safe_call(backend.get().af_transform(ct.pointer(output.arr),

image.arr, trans_mat.arr,

ct.c_longlong(odim0), ct.c_longlong(odim1),

method.value, is_inverse))

return output

def rotate(image, theta, is_crop = True, method = INTERP.NEAREST):

"""

Rotate an image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

theta : scalar

- The angle to rotate in radians.

is_crop : optional: bool. default: True.

- Specifies if the output should be cropped to the input size.

method : optional: af.INTERP. default: af.INTERP.NEAREST.

- Interpolation method used for rotating.

Returns

---------

out : af.Array

- Output image after rotating.

"""

output = Array()

safe_call(backend.get().af_rotate(ct.pointer(output.arr), image.arr,

ct.c_float(theta), is_crop, method.value))

return output

def translate(image, trans0, trans1, odim0 = 0, odim1 = 0, method = INTERP.NEAREST):

"""

Translate an image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

trans0: int.

- Translation along first dimension in pixels.

trans1: int.

- Translation along second dimension in pixels.

odim0 : optional: int. default: 0.

- Size of the first dimension of the output.

odim1 : optional: int. default: 0.

- Size of the second dimension of the output.

method : optional: af.INTERP. default: af.INTERP.NEAREST.

- Interpolation method used for translation.

Returns

---------

out : af.Array

- Output image after translation.

Note

-----

- If `odim0` and `odim` are 0, the output dimensions are automatically calculated by the function.

"""

output = Array()

safe_call(backend.get().af_translate(ct.pointer(output.arr),

image.arr, trans0, trans1,

ct.c_longlong(odim0), ct.c_longlong(odim1), method.value))

return output

def scale(image, scale0, scale1, odim0 = 0, odim1 = 0, method = INTERP.NEAREST):

"""

Scale an image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

scale0 : scalar.

- Scale factor for the first dimension.

scale1 : scalar.

- Scale factor for the second dimension.

odim0 : optional: int. default: None.

- Size of the first dimension of the output.

odim1 : optional: int. default: None.

- Size of the second dimension of the output.

method : optional: af.INTERP. default: af.INTERP.NEAREST.

- Interpolation method used for resizing.

Returns

---------

out : af.Array

- Output image after scaling.

Note

-----

- If `odim0` and `odim` are 0, the output dimensions are automatically calculated by the function.

"""

output = Array()

safe_call(backend.get().af_scale(ct.pointer(output.arr),

image.arr, ct.c_float(scale0), ct.c_float(scale1),

ct.c_longlong(odim0), ct.c_longlong(odim1), method.value))

return output

def skew(image, skew0, skew1, odim0 = 0, odim1 = 0, method = INTERP.NEAREST, is_inverse=True):

"""

Skew an image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

skew0 : scalar.

- Skew factor for the first dimension.

skew1 : scalar.

- Skew factor for the second dimension.

odim0 : optional: int. default: None.

- Size of the first dimension of the output.

odim1 : optional: int. default: None.

- Size of the second dimension of the output.

method : optional: af.INTERP. default: af.INTERP.NEAREST.

- Interpolation method used for resizing.

is_inverse : optional: bool. default: True.

- Specifies if the inverse skew is applied.

Returns

---------

out : af.Array

- Output image after skewing.

Note

-----

- If `odim0` and `odim` are 0, the output dimensions are automatically calculated by the function.

"""

output = Array()

safe_call(backend.get().af_skew(ct.pointer(output.arr),

image.arr, ct.c_float(skew0), ct.c_float(skew1),

ct.c_longlong(odim0), ct.c_longlong(odim1),

method.value, is_inverse))

return output

def histogram(image, nbins, min_val = None, max_val = None):

"""

Find the histogram of an image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

nbins : int.

- Number of bins in the histogram.

min_val : optional: scalar. default: None.

- The lower bound for the bin values.

- If None, `af.min(image)` is used.

max_val : optional: scalar. default: None.

- The upper bound for the bin values.

- If None, `af.max(image)` is used.

Returns

---------

hist : af.Array

- Containing the histogram of the image.

"""

from .algorithm import min as af_min

from .algorithm import max as af_max

if min_val is None:

min_val = af_min(image)

if max_val is None:

max_val = af_max(image)

output = Array()

safe_call(backend.get().af_histogram(ct.pointer(output.arr),

image.arr, ct.c_uint(nbins),

ct.c_float(min_val), ct.c_float(max_val)))

return output

def hist_equal(image, hist):

"""

Equalize an image based on a histogram.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

hist : af.Array

- Containing the histogram of an image.

Returns

---------

output : af.Array

- The equalized image.

"""

output = Array()

safe_call(backend.get().af_hist_equal(ct.pointer(output.arr), image.arr, hist.arr))

return output

def dilate(image, mask = None):

"""

Run image dilate on the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

mask : optional: af.Array. default: None.

- Specifies the neighborhood of a pixel.

- When None, a [3, 3] array of all ones is used.

Returns

---------

output : af.Array

- The dilated image.

"""

if mask is None:

mask = constant(1, 3, 3, dtype=Dtype.f32)

output = Array()

safe_call(backend.get().af_dilate(ct.pointer(output.arr), image.arr, mask.arr))

return output

def dilate3(volume, mask = None):

"""

Run volume dilate on a volume.

Parameters

----------

volume : af.Array

- A 3 D arrayfire array representing a volume, or

- A multi dimensional array representing batch of volumes.

mask : optional: af.Array. default: None.

- Specifies the neighborhood of a pixel.

- When None, a [3, 3, 3] array of all ones is used.

Returns

---------

output : af.Array

- The dilated volume.

"""

if mask is None:

mask = constant(1, 3, 3, 3, dtype=Dtype.f32)

output = Array()

safe_call(backend.get().af_dilate3(ct.pointer(output.arr), volume.arr, mask.arr))

return output

def erode(image, mask = None):

"""

Run image erode on the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

mask : optional: af.Array. default: None.

- Specifies the neighborhood of a pixel.

- When None, a [3, 3] array of all ones is used.

Returns

---------

output : af.Array

- The eroded image.

"""

if mask is None:

mask = constant(1, 3, 3, dtype=Dtype.f32)

output = Array()

safe_call(backend.get().af_erode(ct.pointer(output.arr), image.arr, mask.arr))

return output

def erode3(volume, mask = None):

"""

Run volume erode on the volume.

Parameters

----------

volume : af.Array

- A 3 D arrayfire array representing an volume, or

- A multi dimensional array representing batch of volumes.

mask : optional: af.Array. default: None.

- Specifies the neighborhood of a pixel.

- When None, a [3, 3, 3] array of all ones is used.

Returns

---------

output : af.Array

- The eroded volume.

"""

if mask is None:

mask = constant(1, 3, 3, 3, dtype=Dtype.f32)

output = Array()

safe_call(backend.get().af_erode3(ct.pointer(output.arr), volume.arr, mask.arr))

return output

def bilateral(image, s_sigma, c_sigma, is_color = False):

"""

Apply bilateral filter to the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

s_sigma : scalar.

- Sigma value for the co-ordinate space.

c_sigma : scalar.

- Sigma value for the color space.

is_color : optional: bool. default: False.

- Specifies if the third dimension is 3rd channel (if True) or a batch (if False).

Returns

---------

output : af.Array

- The image after the application of the bilateral filter.

"""

output = Array()

safe_call(backend.get().af_bilateral(ct.pointer(output.arr),

image.arr, ct.c_float(s_sigma),

ct.c_float(c_sigma), is_color))

return output

def mean_shift(image, s_sigma, c_sigma, n_iter, is_color = False):

"""

Apply mean shift to the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

s_sigma : scalar.

- Sigma value for the co-ordinate space.

c_sigma : scalar.

- Sigma value for the color space.

n_iter : int.

- Number of mean shift iterations.

is_color : optional: bool. default: False.

- Specifies if the third dimension is 3rd channel (if True) or a batch (if False).

Returns

---------

output : af.Array

- The image after the application of the meanshift.

"""

output = Array()

safe_call(backend.get().af_mean_shift(ct.pointer(output.arr),

image.arr, ct.c_float(s_sigma), ct.c_float(c_sigma),

ct.c_uint(n_iter), is_color))

return output

def medfilt(image, w0 = 3, w1 = 3, edge_pad = PAD.ZERO):

"""

Apply median filter for the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

w0 : optional: int. default: 3.

- The length of the filter along the first dimension.

w1 : optional: int. default: 3.

- The length of the filter along the second dimension.

edge_pad : optional: af.PAD. default: af.PAD.ZERO

- Flag specifying how the median at the edge should be treated.

Returns

---------

output : af.Array

- The image after median filter is applied.

"""

output = Array()

safe_call(backend.get().af_medfilt(ct.pointer(output.arr),

image.arr, ct.c_longlong(w0),

ct.c_longlong(w1), edge_pad.value))

return output

def minfilt(image, w_len = 3, w_wid = 3, edge_pad = PAD.ZERO):

"""

Apply min filter for the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

w0 : optional: int. default: 3.

- The length of the filter along the first dimension.

w1 : optional: int. default: 3.

- The length of the filter along the second dimension.

edge_pad : optional: af.PAD. default: af.PAD.ZERO

- Flag specifying how the min at the edge should be treated.

Returns

---------

output : af.Array

- The image after min filter is applied.

"""

output = Array()

safe_call(backend.get().af_minfilt(ct.pointer(output.arr),

image.arr, ct.c_longlong(w_len),

ct.c_longlong(w_wid), edge_pad.value))

return output

def maxfilt(image, w_len = 3, w_wid = 3, edge_pad = PAD.ZERO):

"""

Apply max filter for the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

w0 : optional: int. default: 3.

- The length of the filter along the first dimension.

w1 : optional: int. default: 3.

- The length of the filter along the second dimension.

edge_pad : optional: af.PAD. default: af.PAD.ZERO

- Flag specifying how the max at the edge should be treated.

Returns

---------

output : af.Array

- The image after max filter is applied.

"""

output = Array()

safe_call(backend.get().af_maxfilt(ct.pointer(output.arr),

image.arr, ct.c_longlong(w_len),

ct.c_longlong(w_wid), edge_pad.value))

return output

def regions(image, conn = CONNECTIVITY.FOUR, out_type = Dtype.f32):

"""

Find the connected components in the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image.

conn : optional: af.CONNECTIVITY. default: af.CONNECTIVITY.FOUR.

- Specifies the connectivity of the pixels.

out_type : optional: af.Dtype. default: af.Dtype.f32.

- Specifies the type for the output.

Returns

---------

output : af.Array

- An array where each pixel is labeled with its component number.

"""

output = Array()

safe_call(backend.get().af_regions(ct.pointer(output.arr), image.arr,

conn.value, out_type.value))

return output

def sobel_derivatives(image, w_len=3):

"""

Find the sobel derivatives of the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

w_len : optional: int. default: 3.

- The size of the sobel operator.

Returns

---------

(dx, dy) : tuple of af.Arrays.

- `dx` is the sobel derivative along the horizontal direction.

- `dy` is the sobel derivative along the vertical direction.

"""

dx = Array()

dy = Array()

safe_call(backend.get().af_sobel_operator(ct.pointer(dx.arr), ct.pointer(dy.arr),

image.arr, ct.c_uint(w_len)))

return dx,dy

def sobel_filter(image, w_len = 3, is_fast = False):

"""

Apply sobel filter to the image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

w_len : optional: int. default: 3.

- The size of the sobel operator.

is_fast : optional: bool. default: False.

- Specifies if the magnitude is generated using SAD (if True) or SSD (if False).

Returns

---------

output : af.Array

- Image containing the magnitude of the sobel derivatives.

"""

from .arith import abs as af_abs

from .arith import hypot as af_hypot

dx,dy = sobel_derivatives(image, w_len)

if (is_fast):

return af_abs(dx) + af_abs(dy)

else:

return af_hypot(dx, dy)

def rgb2gray(image, r_factor = 0.2126, g_factor = 0.7152, b_factor = 0.0722):

"""

Convert RGB image to Grayscale.

Parameters

----------

image : af.Array

- A 3 D arrayfire array representing an 3 channel image, or

- A multi dimensional array representing batch of images.

r_factor : optional: scalar. default: 0.2126.

- Weight for the red channel.

g_factor : optional: scalar. default: 0.7152.

- Weight for the green channel.

b_factor : optional: scalar. default: 0.0722.

- Weight for the blue channel.

Returns

--------

output : af.Array

- A grayscale image.

"""

output=Array()

safe_call(backend.get().af_rgb2gray(ct.pointer(output.arr),

image.arr, ct.c_float(r_factor), ct.c_float(g_factor), ct.c_float(b_factor)))

return output

def gray2rgb(image, r_factor = 1.0, g_factor = 1.0, b_factor = 1.0):

"""

Convert Grayscale image to an RGB image.

Parameters

----------

image : af.Array

- A 2 D arrayfire array representing an image, or

- A multi dimensional array representing batch of images.

r_factor : optional: scalar. default: 1.0.

- Scale factor for the red channel.

g_factor : optional: scalar. default: 1.0.

- Scale factor for the green channel.

b_factor : optional: scalar. default: 1.0

- Scale factor for the blue channel.

Returns

--------

output : af.Array

- An RGB image.

- The channels are not coalesced, i.e. they appear along the third dimension.

"""

output=Array()

safe_call(backend.get().af_gray2rgb(ct.pointer(output.arr),

image.arr, ct.c_float(r_factor), ct.c_float(g_factor), ct.c_float(b_factor)))

return output

def hsv2rgb(image):

"""

Convert HSV image to RGB.

Parameters

----------

image : af.Array

- A 3 D arrayfire array representing an 3 channel image, or

- A multi dimensional array representing batch of images.

Returns

--------

output : af.Array

- A HSV image.

"""

output = Array()

safe_call(backend.get().af_hsv2rgb(ct.pointer(output.arr), image.arr))

return output

def rgb2hsv(image):

"""

Convert RGB image to HSV.

Parameters

----------

image : af.Array

- A 3 D arrayfire array representing an 3 channel image, or

- A multi dimensional array representing batch of images.

Returns

--------

output : af.Array

- A RGB image.

"""

output = Array()

safe_call(backend.get().af_rgb2hsv(ct.pointer(output.arr), image.arr))

return output

def color_space(image, to_type, from_type):

"""

Convert an image from one color space to another.

Parameters

----------

image : af.Array

- A multi dimensional array representing batch of images in `from_type` color space.

to_type : af.CSPACE

- An enum for the destination color space.

from_type : af.CSPACE

- An enum for the source color space.

Returns

--------

output : af.Array

- An image in the `to_type` color space.

"""

output = Array()

safe_call(backend.get().af_color_space(ct.pointer(output.arr), image.arr,

to_type.value, from_type.value))

return output

def unwrap(image, wx, wy, sx, sy, px=0, py=0, is_column=True):

"""

Unrwap an image into an array.

Parameters

----------

image : af.Array

A multi dimensional array specifying an image or batch of images.

wx : Integer.

Block window size along the first dimension.

wy : Integer.

Block window size along the second dimension.

sx : Integer.

Stride along the first dimension.

sy : Integer.

Stride along the second dimension.

px : Integer. Optional. Default: 0

Padding along the first dimension.

py : Integer. Optional. Default: 0

Padding along the second dimension.

is_column : Boolean. Optional. Default: True.

Specifies if the patch should be laid along row or columns.

Returns

-------

out : af.Array

A multi dimensional array contianing the image patches along specified dimension.

Examples

--------

>>> import arrayfire as af

>>> a = af.randu(6, 6)

>>> af.display(a)

[6 6 1 1]

0.4107 0.3775 0.0901 0.8060 0.0012 0.9250

0.8224 0.3027 0.5933 0.5938 0.8703 0.3063

0.9518 0.6456 0.1098 0.8395 0.5259 0.9313

0.1794 0.5591 0.1046 0.1933 0.1443 0.8684

0.4198 0.6600 0.8827 0.7270 0.3253 0.6592

0.0081 0.0764 0.1647 0.0322 0.5081 0.4387

>>> b = af.unwrap(a, 2, 2, 2, 2)

>>> af.display(b)

[4 9 1 1]

0.4107 0.9518 0.4198 0.0901 0.1098 0.8827 0.0012 0.5259 0.3253

0.8224 0.1794 0.0081 0.5933 0.1046 0.1647 0.8703 0.1443 0.5081

0.3775 0.6456 0.6600 0.8060 0.8395 0.7270 0.9250 0.9313 0.6592

0.3027 0.5591 0.0764 0.5938 0.1933 0.0322 0.3063 0.8684 0.4387

"""

out = Array()

safe_call(backend.get().af_unwrap(ct.pointer(out.arr), image.arr,

ct.c_longlong(wx), ct.c_longlong(wy),

ct.c_longlong(sx), ct.c_longlong(sy),