Correct the algorithm of image filter code

[saoko]

Hi everyone. I'm new in python. Now I'm trying to write an image filter, according to the next algorithm:


i is a row, j is a column, m(i,j) is a pixel, s(i,j) is a sum of pixels, max(m(i,j)) is a max pixel in a row, k is a coefficient (0.7), m is an array of RGB average.

Before using this algorithm I firstly need to convert the image to grayscale. Here a code in python:

import cv2
import numpy as np
import matplotlib.pyplot as plt

img_path = 'image.jpg'

img = cv2.imread(img_path)
imgshape = img.shape

fix_img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

R, G, B = fix_img[:,:,0], fix_img[:,:,1], fix_img[:,:,2]

grayscale_img = np.mean(fix_img, axis=2)

s = np.array([0][0])
b = np.ones(imgshape[:2])
k = 0.7
rows, cols = imgshape[:2] #(192, 184, 3)

s = np.array([0][0])
b = np.ones(imgshape[:2])
k = 0.7
rows, cols = imgshape[:2] #(192, 184, 3)

for j in range(grayscale_img.shape[1]):
    for i in range(grayscale_img.shape[0]):
        max = np.amax(grayscale_img, axis=1)[j]
        m = grayscale_img[j,i]
        s[j,i] = s[j,i] + m
        if s[j,i] >= (k*max):
            s[j,i] = s[j,i] - (k*max)
            s[j,i] = s[j,i] + m
            b[j,i] = 1
        else: 
            s[j,i] = s[j,i] + m
            b[j,i] = 0
cv2.waitKey()
cv2.destroyAllWindows()

While running this code I get the error in line 30

IndexError: too many indices for array: array is 0-dimensional, but 2 were indexed

What is the reason of an error. Could you please help to correct this code?
Thanks!

Here's an example of image filter work:

[deanhystad]

Does this work for you?

rows, cols = imgshape[:2] #(192, 184, 3)
s = np.zeros((rows, cols))
b = np.ones(imgshape[:2])
k = 0.7

[saoko]

Does this work for you?

rows, cols = imgshape[:2] #(192, 184, 3)
s = np.zeros((rows, cols))
b = np.ones(imgshape[:2])
k = 0.7

Thanks, it helped a bit. But I still didn't get the needed result. I made additional corrections (changed indexes backwards in s, m, b). Here's an updated code:

...
rows, cols = imgshape[:2] #(192, 184, 3)
s = np.zeros((rows,cols))
b = np.ones(imgshape[:2])
k = 0.9

for j in range(grayscale_img.shape[1]):
    for i in range(grayscale_img.shape[0]):
        max = np.amax(grayscale_img, axis=1)[j]
        m = grayscale_img[i,j]
        s[i,j] = s[i,j] + m
        if s[i,j] >= (k*max):
            s[i,j] = s[i,j] - (k*max)
            s[i,j] = s[i,j] + m
            b[i,j] = 1
        else: 
            s[i,j] = s[i,j] + m
            b[i,j] = 0

plt.imshow(b, cmap='gray')
plt.savefig('new_img.png')

And this is what I get (the second image is when k=0.7, the third is when k=0.9):


How to correct the code to make image impulse-look-like as in example?

[deanhystad]

Why aren't you letting cv2 do the grayscale conversion?

grayscale = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

And why not use cv2.threshold to convert that to black and white?

[saoko]

Why aren't you letting cv2 do the grayscale conversion?

grayscale = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

And why not use cv2.threshold to convert that to black and white?

It's a part of the task - not to use built-in grayscale filter

[deanhystad]

You should display the grayscale image you produce and the greyscale image that is produced by cv2. That might help you figure out why the thresholding isn't producing the result you want. I don't think using mean is correct. Red and Blue are perceived to be dark colors, and Green a bright color, but in your equation all contribute equally to the brightness of the pixel.

[saoko]

You should display the grayscale image you produce and the greyscale image that is produced by cv2. That might help you figure out why the thresholding isn't producing the result you want. I don't think using mean is correct. Red and Blue are perceived to be dark colors, and Green a bright color, but in your equation all contribute equally to the brightness of the pixel.

I don't think that the problem is in grayscale filter. For comparison,
this is a result of mean grayscale filter


And this is a result of cv2 build-in grayscale filter (cv2.COLOR_BGR2GRAY)


As you see, no so much difference