def guidedfilter(I, p, r, eps): # guidance image: I (should be a gray-scale/single channel image) # filtering imput image: p (should be a gray-scale/single channel image) # local window radius: r # regularization parameter:eps hei, wid = I.shape[: 2] # the size of each local patch N=(2r+1)^2 except for boundary pixels N = bF.boxfilter(np.ones((hei, wid)), r) mean_I = bF.boxfilter(I, r) / N mean_p = bF.boxfilter(p, r) / N mean_Ip = bF.boxfilter(I * p, r) / N # this is the covariance of (I, p) in each local patch cov_Ip = mean_Ip - mean_I * mean_p mean_II = bF.boxfilter( I * I, r) / N var_I = mean_II - mean_I * mean_I # Eqn(5) in the paper a = cov_Ip / (var_I + eps) # Eqn(6) in the paper b = mean_p - a * mean_I mean_a = bF.boxfilter(a, r) / N mean_b = bF.boxfilter(b, r) / N # Eqn(8) in the paper q = mean_a * I + mean_b # return q, N, mean_I, mean_p, mean_Ip, cov_Ip, mean_II, var_I, a, b, mean_a, mean_b, q return q
def guidedfilter(I, p, r, eps):
# guidance image: I (should be a gray-scale/single channel image)
# filtering imput image: p (should be a gray-scale/single channel image)
# local window radius: r
# regularization parameter:eps
hei, wid = I.shape[:2]
# the size of each local patch N=(2r+1)^2 except for boundary pixels
N = bF.boxfilter(np.ones((hei, wid)), r)
mean_I = bF.boxfilter(I, r) / N
mean_p = bF.boxfilter(p, r) / N
mean_Ip = bF.boxfilter(I * p, r) / N
# this is the covariance of (I, p) in each local patch
cov_Ip = mean_Ip - mean_I * mean_p
mean_II = bF.boxfilter(I * I, r) / N
var_I = mean_II - mean_I * mean_I
# Eqn(5) in the paper
a = cov_Ip / (var_I + eps)
# Eqn(6) in the paper
b = mean_p - a * mean_I
mean_a = bF.boxfilter(a, r) / N
mean_b = bF.boxfilter(b, r) / N
# Eqn(8) in the paper
q = mean_a * I + mean_b
# return q, N, mean_I, mean_p, mean_Ip, cov_Ip, mean_II, var_I, a, b, mean_a, mean_b, q
return q
import cv2
import numpy as np
import scipy.io as sio
import boxFilter as bf
img = cv2.imread('./img_enhancement/tulips.bmp') / 255.0
test_img = img[:, :, 0]
r = 16
imDst = bf.boxfilter(test_img, r)
sio.savemat('saveddata.mat', {'imDst_py': imDst, 'test_img': test_img})
def guidedfilter_color(I, p, r, eps): # guidance image: I (should be a color (BGR) image) # filtering input image: p (should be a gray-scale/single channel image) # local window radius: r # regularization parameterL epsilon hei, wid = p.shape # the size of each local patch N=(2r+1)^2 except for boundary pixels N = bF.boxfilter(np.ones((hei, wid)), r) mean_I_r = bF.boxfilter(I[:, :, 2], r) / N mean_I_b = bF.boxfilter(I[:, :, 0], r) / N mean_I_g = bF.boxfilter(I[:, :, 1], r) / N mean_p = bF.boxfilter(p, r) / N mean_Ip_r = bF.boxfilter(I[:, :, 2] * p, r) / N mean_Ip_b = bF.boxfilter(I[:, :, 0] * p, r) / N mean_Ip_g = bF.boxfilter(I[:, :, 1] * p, r) / N # covariance of (I, p) in each local patch cov_Ip_b = mean_Ip_b - mean_I_b * mean_p cov_Ip_g = mean_Ip_g - mean_I_g * mean_p cov_Ip_r = mean_Ip_r - mean_I_r * mean_p # variance of I in each local patch: the matrix Sigma in Eqn(14) # Note the variance in each local patch is a 3*3 symmetric matrix # rr, rg, rb # Sigma = rg, gg, gb # rb, gb, bb var_I_rr = bF.boxfilter(I[:, :, 2] * I[:, :, 2], r) / N - mean_I_r * mean_I_r var_I_rg = bF.boxfilter(I[:, :, 2] * I[:, :, 1], r) / N - mean_I_r * mean_I_g var_I_rb = bF.boxfilter(I[:, :, 2] * I[:, :, 0], r) / N - mean_I_r * mean_I_b var_I_gg = bF.boxfilter(I[:, :, 1] * I[:, :, 1], r) / N - mean_I_g * mean_I_g var_I_gb = bF.boxfilter(I[:, :, 1] * I[:, :, 0], r) / N - mean_I_g * mean_I_b var_I_bb = bF.boxfilter(I[:, :, 0] * I[:, :, 0], r) / N - mean_I_b * mean_I_b a = np.zeros((hei, wid, 3)) for y in range(hei): for x in range(wid): Sigma = np.matrix([[var_I_rr[y, x], var_I_rg[y, x], var_I_rb[y, x]], [var_I_rg[y, x], var_I_gg[y, x], var_I_gb[y, x]], [var_I_rb[y, x], var_I_gb[y, x], var_I_bb[y, x]]]) # Sigma = Sigma + eps * eys(3) cov_Ip = np.matrix([cov_Ip_r[y, x], cov_Ip_g[y, x], cov_Ip_b[y, x]]) # Eqn(14) int the paper a[y, x, :] = cov_Ip * inv(Sigma + eps * np.eye(3)) b = mean_p - a[:, :, 0] * mean_I_r - a[:, :, 1] * mean_I_g - a[:, :, 2] * mean_I_b # Eqn(16) in the paper b1 = bF.boxfilter(a[:, :, 0], r) * I[:, :, 2] q = (bF.boxfilter(a[:, :, 0], r) * I[:, :, 2]\ + bF.boxfilter(a[:, :, 1], r) * I[:, :, 1]\ + bF.boxfilter(a[:, :, 2], r) * I[:, :, 0]\ + bF.boxfilter(b, r)) / N # return q, N, mean_I_b,mean_I_g, mean_I_r,mean_p,mean_Ip_b,mean_Ip_g,mean_Ip_r,cov_Ip_b,cov_Ip_g,cov_Ip_r,var_I_bb,var_I_gb,\ # var_I_rb,var_I_gg,var_I_rg,var_I_rr,a,b,q, b1, cov_Ip return q
import cv2
import numpy as np
import scipy.io as sio
import boxFilter as bf
img = cv2.imread('./img_enhancement/tulips.bmp') / 255.0
test_img = img[:, :, 0]
r = 16
imDst = bf.boxfilter(test_img, r)
sio.savemat('saveddata.mat', {'imDst_py': imDst, 'test_img': test_img})
def guidedfilter_color(I, p, r, eps):
# guidance image: I (should be a color (BGR) image)
# filtering input image: p (should be a gray-scale/single channel image)
# local window radius: r
# regularization parameterL epsilon
hei, wid = p.shape
# the size of each local patch N=(2r+1)^2 except for boundary pixels
N = bF.boxfilter(np.ones((hei, wid)), r)
mean_I_r = bF.boxfilter(I[:, :, 2], r) / N
mean_I_b = bF.boxfilter(I[:, :, 0], r) / N
mean_I_g = bF.boxfilter(I[:, :, 1], r) / N
mean_p = bF.boxfilter(p, r) / N
mean_Ip_r = bF.boxfilter(I[:, :, 2] * p, r) / N
mean_Ip_b = bF.boxfilter(I[:, :, 0] * p, r) / N
mean_Ip_g = bF.boxfilter(I[:, :, 1] * p, r) / N
# covariance of (I, p) in each local patch
cov_Ip_b = mean_Ip_b - mean_I_b * mean_p
cov_Ip_g = mean_Ip_g - mean_I_g * mean_p
cov_Ip_r = mean_Ip_r - mean_I_r * mean_p
# variance of I in each local patch: the matrix Sigma in Eqn(14)
# Note the variance in each local patch is a 3*3 symmetric matrix
# rr, rg, rb
# Sigma = rg, gg, gb
# rb, gb, bb
var_I_rr = bF.boxfilter(I[:, :, 2] * I[:, :, 2],
r) / N - mean_I_r * mean_I_r
var_I_rg = bF.boxfilter(I[:, :, 2] * I[:, :, 1],
r) / N - mean_I_r * mean_I_g
var_I_rb = bF.boxfilter(I[:, :, 2] * I[:, :, 0],
r) / N - mean_I_r * mean_I_b
var_I_gg = bF.boxfilter(I[:, :, 1] * I[:, :, 1],
r) / N - mean_I_g * mean_I_g
var_I_gb = bF.boxfilter(I[:, :, 1] * I[:, :, 0],
r) / N - mean_I_g * mean_I_b
var_I_bb = bF.boxfilter(I[:, :, 0] * I[:, :, 0],
r) / N - mean_I_b * mean_I_b
a = np.zeros((hei, wid, 3))
for y in range(hei):
for x in range(wid):
Sigma = np.matrix(
[[var_I_rr[y, x], var_I_rg[y, x], var_I_rb[y, x]],
[var_I_rg[y, x], var_I_gg[y, x], var_I_gb[y, x]],
[var_I_rb[y, x], var_I_gb[y, x], var_I_bb[y, x]]])
# Sigma = Sigma + eps * eys(3)
cov_Ip = np.matrix(
[cov_Ip_r[y, x], cov_Ip_g[y, x], cov_Ip_b[y, x]])
# Eqn(14) int the paper
a[y, x, :] = cov_Ip * inv(Sigma + eps * np.eye(3))
b = mean_p - a[:, :, 0] * mean_I_r - a[:, :,
1] * mean_I_g - a[:, :,
2] * mean_I_b
# Eqn(16) in the paper
b1 = bF.boxfilter(a[:, :, 0], r) * I[:, :, 2]
q = (bF.boxfilter(a[:, :, 0], r) * I[:, :, 2]\
+ bF.boxfilter(a[:, :, 1], r) * I[:, :, 1]\
+ bF.boxfilter(a[:, :, 2], r) * I[:, :, 0]\
+ bF.boxfilter(b, r)) / N
# return q, N, mean_I_b,mean_I_g, mean_I_r,mean_p,mean_Ip_b,mean_Ip_g,mean_Ip_r,cov_Ip_b,cov_Ip_g,cov_Ip_r,var_I_bb,var_I_gb,\
# var_I_rb,var_I_gg,var_I_rg,var_I_rr,a,b,q, b1, cov_Ip
return q
