Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

3x Speed-up #1

Merged
merged 2 commits into from
Jun 24, 2013
+20 −16
Merged

3x Speed-up #1

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
47c5e91
Resolving conflict from cherry-picking 82597a3
w-m Jun 23, 2013
fa23afc
using -= where appropriate
w-m Jun 23, 2013
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 20 additions & 16 deletions ssim/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@

import numpy
import scipy.ndimage
from numpy.ma.core import exp
from numpy.ma.core import exp, sqrt
from scipy.constants.constants import pi
from compat import ImageOps

Expand All @@ -30,6 +30,11 @@ def _to_grayscale(im):

return gray, alpha

def convolve_gaussian_2d(image, gaussian_kernel_1d):
result = scipy.ndimage.filters.correlate1d(image, gaussian_kernel_1d, axis = 0)
result = scipy.ndimage.filters.correlate1d(result, gaussian_kernel_1d, axis = 1)
return result

def compute_ssim(im1, im2, gaussian_kernel_sigma=1.5, gaussian_kernel_width=11):
"""
The function to compute SSIM
Expand All @@ -38,15 +43,14 @@ def compute_ssim(im1, im2, gaussian_kernel_sigma=1.5, gaussian_kernel_width=11):
@return: SSIM float value
"""

#Gaussian kernel definition
gaussian_kernel = numpy.zeros((gaussian_kernel_width, gaussian_kernel_width))

# 1D Gaussian kernel definition
gaussian_kernel_1d = numpy.ndarray((gaussian_kernel_width))
mu = int(gaussian_kernel_width / 2)

#Fill Gaussian kernel
for i in range(gaussian_kernel_width):
for j in range(gaussian_kernel_width):
gaussian_kernel[i, j] = \
(1 / (2 * pi * (gaussian_kernel_sigma ** 2))) * \
exp(-(((i - 5) ** 2) + ((j - 5) ** 2)) / (2 * (gaussian_kernel_sigma ** 2)))
for i in xrange(gaussian_kernel_width):
gaussian_kernel_1d[i] = (1 / (sqrt(2 * pi) * (gaussian_kernel_sigma))) * \
exp(-(((i - mu) ** 2)) / (2 * (gaussian_kernel_sigma ** 2)))

# convert the images to grayscale
img_mat_1, img_alpha_1 = _to_grayscale(im1)
Expand All @@ -63,24 +67,24 @@ def compute_ssim(im1, im2, gaussian_kernel_sigma=1.5, gaussian_kernel_width=11):
img_mat_12 = img_mat_1 * img_mat_2

#Means obtained by Gaussian filtering of inputs
img_mat_mu_1 = scipy.ndimage.filters.convolve(img_mat_1, gaussian_kernel)
img_mat_mu_2 = scipy.ndimage.filters.convolve(img_mat_2, gaussian_kernel)
img_mat_mu_1 = convolve_gaussian_2d(img_mat_1, gaussian_kernel_1d)
img_mat_mu_2 = convolve_gaussian_2d(img_mat_2, gaussian_kernel_1d)

#Squares of means
img_mat_mu_1_sq = img_mat_mu_1 ** 2
img_mat_mu_2_sq = img_mat_mu_2 ** 2
img_mat_mu_12 = img_mat_mu_1 * img_mat_mu_2

#Variances obtained by Gaussian filtering of inputs' squares
img_mat_sigma_1_sq = scipy.ndimage.filters.convolve(img_mat_1_sq, gaussian_kernel)
img_mat_sigma_2_sq = scipy.ndimage.filters.convolve(img_mat_2_sq, gaussian_kernel)
img_mat_sigma_1_sq = convolve_gaussian_2d(img_mat_1_sq, gaussian_kernel_1d)
img_mat_sigma_2_sq = convolve_gaussian_2d(img_mat_2_sq, gaussian_kernel_1d)

#Covariance
img_mat_sigma_12 = scipy.ndimage.filters.convolve(img_mat_12, gaussian_kernel)
img_mat_sigma_12 = convolve_gaussian_2d(img_mat_12, gaussian_kernel_1d)

#Centered squares of variances
img_mat_sigma_1_sq = img_mat_sigma_1_sq - img_mat_mu_1_sq
img_mat_sigma_2_sq = img_mat_sigma_2_sq - img_mat_mu_2_sq
img_mat_sigma_1_sq -= img_mat_mu_1_sq
img_mat_sigma_2_sq -= img_mat_mu_2_sq
img_mat_sigma_12 = img_mat_sigma_12 - img_mat_mu_12

#set k1,k2 & c1,c2 to depend on L (width of color map)
Expand Down