Bayer RAW#

Tip

Learn more: https://en.wikipedia.org/wiki/Bayer_filter

Convert between a one-channel Bayer mosaic (*, 1, H, W) and an RGB image. The CFA enum gives the pattern of the mosaic.

Functions#

kornia.color.rgb_to_raw(image, cfa)[source]#

Sample an RGB image into a Bayer raw image.

Try in browser

raw = rgb_to_raw(image, cfa) keeps one channel at each pixel of a (*, 3, H, W) RGB image, the one that the pattern cfa puts there, and returns the (*, 1, H, W) mosaic.

Convention:: The inverse conversion requires even height and width.

Parameters:

image (Tensor) – RGB image with shape (*, 3, H, W).
cfa (CFA) – pattern of the mosaic, see CFA.

Return type:

Tensor

Returns:

raw mosaic with shape (*, 1, H, W).

See also

raw_to_rgb(): the inverse, which interpolates the missing samples.
RgbToRaw: the module form.

Example

>>> rgbinput = torch.rand(2, 3, 4, 6)
>>> raw = rgb_to_raw(rgbinput, CFA.BG) # 2x1x4x6

kornia.color.raw_to_rgb(image, cfa)[source]#

Demosaic a Bayer raw image to RGB.

rgb = raw_to_rgb(image, cfa) converts a (*, 1, H, W) mosaic in \([0, 1]\) to a full-resolution (*, 3, H, W) RGB image. It interpolates red and blue bilinearly and green from its four nearest green samples.

Convention:: Height and width must be even. The mosaic samples are kept, so rgb_to_raw(raw_to_rgb(x, cfa), cfa) returns x. Away from the outermost pixel ring the result equals OpenCV’s COLOR_BAYER_*2RGB up to its 8-bit rounding (checked against OpenCV 5.0). In that ring OpenCV repeats the values of the ring inside it, and this function keeps the samples.

Parameters:

image (Tensor) – raw mosaic with shape (*, 1, H, W).
cfa (CFA) – pattern of the mosaic, see CFA.

Return type:

Tensor

Returns:

RGB image with shape (*, 3, H, W).

See also

raw_to_rgb_2x2_downscaled(): one RGB pixel per 2x2 cell, without interpolation.
rgb_to_raw(): samples an RGB image back into a mosaic.
RawToRgb: the module form.

Example

>>> rawinput = torch.randn(2, 1, 4, 6)
>>> rgb = raw_to_rgb(rawinput, CFA.RG) # 2x3x4x6

kornia.color.raw_to_rgb_2x2_downscaled(image, cfa)[source]#

Convert a Bayer raw image to RGB at half the height and width.

rgb = raw_to_rgb_2x2_downscaled(image, cfa) turns each 2x2 cell of a (*, 1, H, W) mosaic in \([0, 1]\) into one RGB pixel: red and blue are the cell’s samples and green is the mean of its two green samples.

Convention:: Height and width must be even.

Parameters:

image (Tensor) – raw mosaic with shape (*, 1, H, W).
cfa (CFA) – pattern of the mosaic, see CFA.

Return type:

Tensor

Returns:

RGB image with shape (*, 3, H / 2, W / 2).

See also

raw_to_rgb(): keeps the resolution and interpolates.
RawToRgb2x2Downscaled: the module form.

Example

>>> rawinput = torch.randn(2, 1, 4, 6)
>>> rgb = raw_to_rgb_2x2_downscaled(rawinput, CFA.RG) # 2x3x2x3

Modules#

class kornia.color.CFA(value)[source]#

Pattern of the Bayer color filter array of a raw image.

CFA.BG and its siblings name the 2x2 cell that tiles the mosaic. As in OpenCV, the name reads pixel (1, 1) and its right neighbor, so BG is the sensor usually called RGGB. The members, with the colors of the top-left cell read row by row:

BG: R G / G B.
GB: G R / B G.
RG: B G / G R.
GR: G B / R G.

Rotating or flipping a mosaic of even height and width changes its pattern, and so does cropping an odd number of rows or columns from its top or left.

See also

raw_to_rgb(): demosaics a mosaic with this pattern.
rgb_to_raw(): samples an RGB image with this pattern.

References

[1] https://en.wikipedia.org/wiki/Color_filter_array

BG = 0#

GB = 1#

GR = 3#

RG = 2#

class kornia.color.RgbToRaw(cfa)[source]#

Sample an RGB image into a Bayer raw image (module form).

Try in browser

conv = RgbToRaw(cfa) stores the pattern, and out = conv(image) is rgb_to_raw() with it.

Convention:

See the convention block of rgb_to_raw().

Shape:

image: (*, 3, H, W)
output: (*, 1, H, W)

See also

rgb_to_raw(): the one-call form.
RawToRgb: the inverse.

Example

>>> rgbinput = torch.rand(2, 3, 4, 6)
>>> raw = RgbToRaw(CFA.GB)
>>> output = raw(rgbinput)  # 2x1x4x6

References

[1] https://docs.opencv.org/4.0.1/de/d25/imgproc_color_conversions.html

class kornia.color.RawToRgb(cfa)[source]#

Demosaic a Bayer raw image to RGB (module form).

conv = RawToRgb(cfa) stores the pattern, and out = conv(image) is raw_to_rgb() with it.

Convention:

See the convention block of raw_to_rgb().

Shape:

image: (*, 1, H, W)
output: (*, 3, H, W)

See also

raw_to_rgb(): the one-call form.
RawToRgb2x2Downscaled: half the resolution, without interpolation.
RgbToRaw: the inverse.

Example

>>> rawinput = torch.rand(2, 1, 4, 6)
>>> rgb = RawToRgb(CFA.RG)
>>> output = rgb(rawinput)  # 2x3x4x6

class kornia.color.RawToRgb2x2Downscaled(cfa)[source]#

Convert a Bayer raw image to RGB at half the height and width (module form).

conv = RawToRgb2x2Downscaled(cfa) stores the pattern, and out = conv(image) is raw_to_rgb_2x2_downscaled() with it.

Convention:

See the convention block of raw_to_rgb_2x2_downscaled().

Shape:

image: (*, 1, H, W)
output: (*, 3, H / 2, W / 2)

See also

raw_to_rgb_2x2_downscaled(): the one-call form.
RawToRgb: keeps the resolution and interpolates.

Example

>>> rawinput = torch.rand(2, 1, 4, 6)
>>> rgb_downscale = RawToRgb2x2Downscaled(CFA.RG)
>>> output = rgb_downscale(rawinput)  # 2x3x2x3