Skip to content

SynthShapes.hist

MatchHistogram 

MatchHistogram(mean=0.0, std=0.2, num_bins=256)

Bases: Module

Histogram Matching Module to map the intensity values of an image to follow a normal distribution.

PARAMETER DESCRIPTION
mean

Mean of the normal distribution, by default 0.0.

TYPE: float DEFAULT: 0.0

std

Standard deviation of the normal distribution, by default 0.2.

TYPE: float DEFAULT: 0.2

num_bins

Number of bins for histogram, by default 256.

TYPE: int DEFAULT: 256

Source code in SynthShapes/hist.py 
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
def __init__(self, mean=0.0, std=0.2, num_bins=256):
    """
    Histogram Matching Module to map the intensity values of an image to
    follow a normal distribution.

    Parameters
    ----------
    mean : float, optional
        Mean of the normal distribution, by default 0.0.
    std : float, optional
        Standard deviation of the normal distribution, by default 0.2.
    num_bins : int, optional
        Number of bins for histogram, by default 256.
    """
    super(MatchHistogram, self).__init__()
    self.mean = mean
    self.std = std
    self.num_bins = num_bins

calculate_cdf 

calculate_cdf(hist)

Calculate the cumulative distribution function (CDF) for a histogram.

Source code in SynthShapes/hist.py 
30
31
32
33
34
35
def calculate_cdf(self, hist):
    """Calculate the cumulative distribution function (CDF) for a
    histogram."""
    cdf = hist.cumsum(0)
    cdf_normalized = cdf / cdf[-1]
    return cdf_normalized

forward 

forward(source)

Forward pass to perform histogram matching.

PARAMETER DESCRIPTION
source

Source image (HxW), normalized between -1 and 1.

TYPE: Tensor

RETURNS DESCRIPTION
matched

The transformed source image with histogram matching a normal distribution.

TYPE: Tensor

Source code in SynthShapes/hist.py 
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
def forward(self, source):
    """
    Forward pass to perform histogram matching.

    Parameters
    ----------
    source : torch.Tensor
        Source image (HxW), normalized between -1 and 1.

    Returns
    -------
    matched : torch.Tensor
        The transformed source image with histogram matching a normal
        distribution.
    """
    device = source.device

    # Normalize the source image to the range [0, 255] for histogram
    # computation
    source_normalized = ((source + 1) / 2 * 255).clamp(0, 255).long()

    # Compute the histogram and CDF of the source image
    src_hist = torch.histc(source_normalized.float(),
                           bins=self.num_bins, min=0, max=255).to(device)
    src_cdf = self.calculate_cdf(src_hist)

    # Create the normal distribution CDF
    normal_values = torch.linspace(-1, 1, self.num_bins, device=device)
    normal_cdf = torch.distributions.Normal(
        self.mean, self.std).cdf(normal_values)
    normal_cdf = normal_cdf / normal_cdf[-1]  # Normalize to range [0, 1]

    # Create a lookup table to map the pixel values
    lookup_table = torch.zeros(self.num_bins, device=device)
    for src_pixel in range(self.num_bins):
        normal_pixel = torch.searchsorted(normal_cdf, src_cdf[src_pixel])
        lookup_table[src_pixel] = normal_pixel

    # Apply the lookup table to the source image
    source_flat = source_normalized.flatten().long()
    matched_flat = lookup_table[source_flat]
    matched = matched_flat.view(source.shape).float()

    # Convert matched image back to the range [-1, 1]
    matched = matched / (self.num_bins - 1) * 2 - 1

    return matched