Error trying to manually create Gaussian Filter in Python-CodePudding

I created a Python code based on a MATLAB code I already had to implement the Gaussian Filter. But when I run the project, I get the following error:

Traceback (most recent call last):
  File "C:\Users\gisla\Scilab_PDI\gaussianFilter.py", line 37, in <module>
    conv = np.multiply(temp, mascara);
ValueError: operands could not be broadcast together with shapes (4,4,7) (5,5)

The code is this below. The image 'result.jpg' is simply a color image that I am trying to filter on. It has no noise because it was just testing if the code would actually modify the image.

import cv2
import matplotlib as plt
from math import e;

img = cv2.imread("result.jpg");
I = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

sigma = 1;
# Máscara 5x5
mascara = np.zeros((5,5));
W = 0;

def getExp(n):
  return e**n

#Filter
for i in range (1, 5):
    for j in range (1, 5):
        sq_dist = (i-3)^2 (j-3)^2;
        mascara[i,j] = getExp(-1*(sq_dist)/(2*sigma*sigma));
        W = W   mascara[i,j];

mascara = mascara/W;

[m,n] = np.shape(I);
saida = np.zeros((m,n), dtype=np.uint8);
Im = np.pad(img, [2,2]);

for i in range(1, m):
    for j in range(1, n):
        temp = Im[i:i 4, j:j 4];
        temp = np.double(temp);
        conv = np.multiply(temp, mascara);
        saida[i,j] = np.sum(conv[:])


cv2.imwrite('resultado.jpg', img)

I don't know much about Python, so the changes made so far didn't solve the problem.

CodePudding user response：

Python has "Zero-based numbering" so your Mask currently looks like this:

You can see that the first row and column are empty and that you need to start (probably all 4) for-loops from 0.

The second problem is using "^" (XOR operator) for squaring (when calculation "sq_dist"). Use "**" instead.

Use:

import matplotlib.pyplot as plt
plt.imshow(mascara, "gray")

to visualize your kernel.

You also get the multiplication error from beginning the third and fourth for-loop from 1. Also don't use semi-columns at the end of line in python :)

CodePudding user response：

In the line Im = np.pad(img, [2,2]) you use the padding on the img variable, which is your color image with shape (X, Y, 3) (3 dimensions for R, G and B). If greyscale images is what you want to work with, I think instead you probably want to use your I variable with shape (X, Y), so Im = np.pad(I, [2,2]).

Also please also note that Python indexing works different than Matlab indexing. Indexing starts at 0, you have to change the lines

for i in range(1, m):
    for j in range(1, n):
        temp = Im[i:i 4, j:j 4]

for i in range(0, m):
    for j in range(0, n):
        temp = Im[i:i 5, j:j 5]