"""
Full-reference Image Quality Assessment (FR-IQA) methods
"""
# main imports
import numpy as np
# computations imports
from skimage.measure import compare_ssim, compare_psnr
# ipfml imports
from ipfml.exceptions import NumpyShapeComparisonException
def _prepare_arrays(img_true, img_test, p_dtype='float32'):
"""
Prepare image data
Raises:
NumpyShapeComparisonException: if shape of images are not the same
"""
img_true = np.asarray(img_true, dtype=p_dtype)
img_test = np.asarray(img_test, dtype=p_dtype)
if img_true.shape != img_test.shape:
raise NumpyShapeComparisonException
return img_true, img_test
[docs]def mse(img_true, img_test):
"""Returns Mean-Squared Error score between two Numpy arrays
Args:
img_true: Image, numpy array of any dimension
img_test: Image, numpy array of any dimension
Returns:
Computed MSE score
Raises:
NumpyShapeComparisonException: if shape of images are not the same
Example:
>>> from ipfml.iqa import fr
>>> import numpy as np
>>> arr1 = np.arange(10)
>>> arr2 = np.arange(5, 15)
>>> mse_score = fr.mse(arr1, arr2)
>>> mse_score
25.0
"""
img_true, img_test = _prepare_arrays(img_true, img_test)
return np.mean(np.square(img_true - img_test), dtype=np.float64)
[docs]def rmse(img_true, img_test):
"""Returns Root Mean-Squared Error score between two Numpy arrays
Args:
img_true: Image, numpy array of any dimension
img_test: Image, numpy array of any dimension
Returns:
Computed RMSE score
Raises:
NumpyShapeComparisonException: if shape of images are not the same
Example:
>>> from ipfml.iqa import fr
>>> import numpy as np
>>> arr1 = np.arange(10)
>>> arr2 = np.arange(5, 15)
>>> rmse_score = fr.rmse(arr1, arr2)
>>> rmse_score
5.0
"""
return np.sqrt(mse(img_true, img_test))
[docs]def mae(img_true, img_test):
"""Returns Mean Absolute Error between two Numpy arrays
Args:
img_true: Image, numpy array of any dimension
img_test: Image, numpy array of any dimension
Returns:
Computed MAE score
Raises:
NumpyShapeComparisonException: if shape of images are not the same
Example:
>>> from ipfml.iqa import fr
>>> import numpy as np
>>> arr1 = np.arange(10)
>>> arr2 = np.arange(5, 15)
>>> mae_score = fr.mae(arr1, arr2)
>>> mae_score
5.0
"""
img_true, img_test = _prepare_arrays(img_true, img_test)
return np.mean(np.absolute(img_true - img_test), dtype=np.float64)
[docs]def psnr(img_true, img_test):
"""Returns the computed Peak Signal to Noise Ratio (PSNR) between two images
Args:
img_true: Image, numpy array of any dimension
img_test: Image, numpy array of any dimension
Returns:
Computed PSNR score
Example:
>>> from ipfml.iqa import fr
>>> import numpy as np
>>> arr1 = np.arange(100).reshape(10, 10)
>>> arr2 = np.arange(5, 105).reshape(10, 10)
>>> psnr_score = fr.psnr(arr1, arr2)
>>> int(psnr_score)
34
"""
img_true, img_test = _prepare_arrays(img_true, img_test, 'uint8')
return compare_psnr(img_true, img_test)
[docs]def ssim(img_true, img_test):
"""Returns the computed Structural Similarity (SSIM) between two images
Args:
img_true: Image, numpy array of any dimension
img_test: Image, numpy array of any dimension
Returns:
Computed SSIM score
Example:
>>> from ipfml.iqa import fr
>>> import numpy as np
>>> arr1 = np.arange(100).reshape(10, 10)
>>> arr2 = np.arange(5, 105).reshape(10, 10)
>>> ssim_score = fr.ssim(arr1, arr2)
>>> int(ssim_score)
0
"""
img_true, img_test = _prepare_arrays(img_true, img_test)
if img_true.ndim == 3:
return compare_ssim(img_true, img_test, multichannel=True)
else:
return compare_ssim(img_true, img_test)
[docs]def ms_ssim(img_true, img_test):
"""
Implemented later..
"""
pass
[docs]def vif(img_true, img_test):
"""
Implemented later..
"""
pass