I am getting black Laplacian of Guassian image

Question

import cv2
import math

gaussian function

def gaussianblur(img,sigma):
      if(sigma<0):
        print("SIGMA SHOULD BE POSITIVE")
        return;

calculating 1 dimensional kernal with g(x)=(1/squareroot(2*sigma*sigma*3.142)) * e^(-(x*x)/(2*sigma*sigma))

      deno=(((math.sqrt(2*3.142*sigma*sigma))))
      k=[0,0,0,0,0]
      sum=0




      for x in range(-2,3):
       numo=(math.exp(-((x*x)/(2*(sigma*sigma)))))
       k[x+2]=(numo/deno)
       sum=sum+k[x+2]




      for x in range(0,5):
        k[x]=(k[x]/sum)

applying convolution row by row

      for i in range(0,img.shape[0]):
       for j in range(2,img.shape[1]-2):
            img[i,j]=abs((img[i,j-2]*k[0])+(img[i,j-1]*k[1])+(img[i,j]*k[2])+(img[i,j+1]*k[3])+(img[i,j+2]*k[4]))



      return img;  `#end of gaussian blur function`

main function start

read image

dog=img = cv2.imread('art.jpg',cv2.IMREAD_GRAYSCALE)

apply 1 st blur

temp=img=gaussianblur(img,1)

#display image
cv2.imshow('blur1',img)

apply 2nd blur

temp=gaussianblur(temp,1)

cv2.imshow('blur2',temp)

the difference of Gaussian

for i in range(0,img.shape[0]):
    for j in range(0,img.shape[1]):
       dog[i,j]=abs((img[i,j])-(temp[i,j]))

cv2.imshow('DoG',dog)

output

blur1

blur2

DoG

you can refer http://www.tjscientific.com/2017/01/31/using-python-and-opencv-to-create-a-difference-of-gaussian-filter/ — Sanket, Nov 01 '17 at 07:15
actually, i wanted to implement Gaussian blur without built-in function. — abhishek satmar, Nov 01 '17 at 09:00
Try calling `normalize()` prior to displaying your image in case the values are diminishingly small. — Mark Setchell, Nov 01 '17 at 10:23

score 1 · Accepted Answer · answered Nov 02 '17 at 04:10

1

You are overwriting your input here:

for i in range(0,img.shape[0]):
    for j in range(2,img.shape[1]-2):
        img[i,j]=abs((img[i,j-2]*k[0])+(img[i,j-1]*k[1])+(img[i,j]*k[2])+(img[i,j+1]*k[3])+(img[i,j+2]*k[4]))

Try writing the result to a new image.

I do not know how the OpenCV python interface works, does temp=img cause temp to share the data with img (as in, when you change the one you also change the other)? Make sure you have two different data blocks!

answered Nov 02 '17 at 04:10

Cris Luengo

55,762
10
62
120

1

temp = np.zeros((img.shape[0],img.shape[1],1), np.uint8) to create a new empty image with same dimension as your original image. – flamelite Nov 02 '17 at 04:50
I got the result, that was very helpful, thank u @ Cris Luengo – abhishek satmar Nov 02 '17 at 14:41

I am getting black Laplacian of Guassian image

1 Answers1