I have two filters: (1) Median Filter and (2) Adaptive Median Filter
My median filter works. It takes an image, creates a padding the kernel convolves, replacing the center pixel with the median value found in each kernel window.
My adaptive median filter does not work. The difference between this one and the first one is that I want my adaptive median filter to iterate through the image block by block, instead of pixel by pixel, replacing the outliers in each kernel window with the median value. I think my logic is correct in my adaptive median, but I'm having trouble lining everything up in my output, especially with the dimensions of the image requiring the padding.
Also, is there a way to not include the padding when calculating the median near the edges?
Had to change the main for loop:
My median filter works. It takes an image, creates a padding the kernel convolves, replacing the center pixel with the median value found in each kernel window.
My adaptive median filter does not work. The difference between this one and the first one is that I want my adaptive median filter to iterate through the image block by block, instead of pixel by pixel, replacing the outliers in each kernel window with the median value. I think my logic is correct in my adaptive median, but I'm having trouble lining everything up in my output, especially with the dimensions of the image requiring the padding.
Also, is there a way to not include the padding when calculating the median near the edges?
import cv2
import numpy as np
from matplotlib import pyplot as plt
import math
import sys
from skimage.exposure import rescale_intensity
np.set_printoptions(threshold=sys.maxsize)
# PART 2: ADAPTIVE MEDIAN FILTER
# load image
image = cv2.imread('2_noise.png',0)
image2 = cv2.imread('2_noise.png',0)
# convolve function
def medianFilter(image, kernel):
# size of image and kernel
(iH, iW) = image.shape[:2]
(kH, kW) = kernel.shape[:2]
print(iH, iW, kH, kW)
# add padding to input image
pad = (kW - 1) // 2
image = cv2.copyMakeBorder(image, pad, pad, pad, pad,
cv2.BORDER_REPLICATE)
# allocate memory for the output image
output = np.zeros((iH, iW), dtype="float32")
# loop over the input image
for y in np.arange(pad, iH + pad):
for x in np.arange(pad, iW + pad):
# extract the *center* region of the current (x, y)-coordinates dimensions
roi = image[y - pad:y + pad + 1, x - pad:x + pad + 1]
# convolution
k = np.median(roi*kernel)
# store the convolved value in the output(x,y)
output[y - pad, x - pad] = k
# rescale the output image
output = rescale_intensity(output, in_range=(0, 255))
output = (output * 255).astype("uint8")
# return the output image
return output
def iqr(a, outlierConstant):
"""
a : numpy.ndarray (array from which outliers have to be removed.)
outlierConstant : (scale factor around interquartile region.)
"""
num = a.shape[0]
upper_quartile = np.percentile(a, 75)
lower_quartile = np.percentile(a, 25)
IQR = (upper_quartile - lower_quartile) * outlierConstant
quartileSet = (lower_quartile - IQR, upper_quartile + IQR)
outlier_indx = []
for i in range(num):
if np.any(a[i] >= quartileSet[0]) and np.any(a[i] <= quartileSet[1]): pass
else: outlier_indx += [i]
return outlier_indx
def function(arrayMatrix, threshold=.5):
zscore = (arrayMatrix - arrayMatrix.mean())/arrayMatrix.std()
return np.where(np.abs(zscore) > threshold)
def adaptiveMedian(image, kernel):
# size of image and kernel
(iH, iW) = image.shape[:2]
(kH, kW) = kernel.shape[:2]
print('Image Size: ', iH, iW)
print('Kernel Size: ', kH, kW)
# add padding to input image
pad = (kW - 1) // 2
image = cv2.copyMakeBorder(image, pad, pad, pad, pad,
cv2.BORDER_REPLICATE)
# allocate memory for the output image
output = np.zeros((iH+pad*2, iW+pad*2), dtype="float32")
yRange = np.arange(pad, iH + pad)
for xxx in yRange[::7]:
print(xxx)
# loop over the input image
for y in yRange[::7]:
for x in yRange[::7]:
#print(np.arange(pad, iH + pad))
# extract the *center* region of the current (x, y)-coordinates dimensions
roi = image[y - pad:y + pad + 1, x - pad:x + pad + 1]
#print('roi: ', roi)
# finding median
median = np.median(roi)
# finding outliers & reshaping
outlier = function(roi)
out = np.asarray(outlier,order=2)
rowss = out.shape[0]
colss = out.shape[1]
xx = out[0]
yy = out[1]
data = np.array((xx,yy)).T
# replacing outlier pixels with median value
Range = len(data)
for xx in range(0,Range):
xCoord = data[xx][0]
yCoord = data[xx][1]
roi[xCoord][yCoord] = median
#print('roi/median: ', roi)
# convolution
k = (roi*kernel)
#print('k: ', k)
# store refined mask into output image (removing salt & pepper)
#print('y: ', y)
#print('pad: ', pad)
#print('x: ', x)
output[y - pad:y + pad + 1, x - pad:x + pad + 1] = k
# rescale the output image
output = rescale_intensity(output, in_range=(0, 255))
output = (output * 255).astype("uint8")
# return the output image
return output
# construct kernel mask
median = np.ones((7, 7), dtype="float")
# run the functions
adaptiveOutput = adaptiveMedian(image, median)
medianOutput = medianFilter(image2, median)
# plot both images
plt.figure(figsize=(11,6))
plt.subplot(131),plt.imshow(image, cmap = 'gray')
plt.title('Original Image w/ noise added'), plt.xticks([]), plt.yticks([])
plt.subplot(132),plt.imshow(medianOutput, cmap = 'gray')
plt.title('Median Filter'), plt.xticks([]), plt.yticks([])
plt.subplot(133),plt.imshow(adaptiveOutput, cmap = 'gray')
plt.title('Adaptive Median Filter'), plt.xticks([]), plt.yticks([])
plt.show()Had to change the main for loop:
yRange = np.arange(pad, iH + pad)
xRange = np.arange(pad, iW + pad)
# loop over the input image
for y in yRange[::6]:
for x in xRange[::6]:Also putting the output[...] = k line in the for xx loop, everything seems to work now: for xx in range(0,Range):
xCoord = data[xx][0]
yCoord = data[xx][1]
roi[xCoord][yCoord] = median
#print('roi/median: ', roi)
# convolution
k = (roi*kernel)
#print('k: ', k)
# store refined mask into output image (removing salt & pepper)
#print('y: ', y)
#print('pad: ', pad)
#print('x: ', x)
output[y - pad:y + pad + 1, x - pad:x + pad + 1] = k