I have a binary image of large size (2000x2000). In this image most of the pixel values are zero and some of them are 1. I need to get only 100 randomly chosen pixel coordinates with value 1 from image. I am beginner in python, so please answer.
4 Answers
I'd suggest making a list of coordinates of all non-zero pixels (by checking all pixels in the image), then using random.shuffle on the list and taking the first 100 elements.
4 Comments
technusm1
just a quick question - will he need to be careful about 2080 limit as mentioned here?: docs.python.org/3/library/random.html#random.shuffle
Błotosmętek
A valid question. I guess it is up to OP to decide, because we don't know what "most" in the question means - 90%? 99%? 99.99%? If there are just a few hundred non-zero pixels, it should be safe to use
random.shuffle, otherwise it might be better to use random.choice.
vishnu saini
Awesome..! Thankyou sir. I found that random.choice can not be used here as coordinates are 2 dimensional. so i have tried random.shuffle as- coordinates = np.argwhere(img) print(coordinates.shape) index = random.shuffle(coordinates) print(index[0:100]) it gives me following error-TypeError: 'NoneType' object is not subscriptable
vishnu saini
coordinates = np.argwhere(img) print(coordinates.shape) random.shuffle(coordinates) print(coordinates[0:100])After importing necessary libraries like
import cv2
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
gray_img = cv2.imread(img_file, cv2.IMREAD_GRAYSCALE) # grayscale
gray_img[i,j] will give pixel value at (i,j) position
Try to send all these values into a file in this format
i_positition,j_position,value_of_pixel
path = os.getcwd() + '/filename.txt'
data = pd.read_csv(path, header=None, names=['i', 'j', 'value'])
positive = data[data['value'].isin([1])]
negative = data[data['value'].isin([0])]
positive data frame contains all the pixel positions whose value is 1.
positive['i'] ,positive['j'] will give you list of (i,j) values of all the pixels whose value is 1.
i_val=np.asarray(positive['i'])
j_val=np.asarray(positive['j'])
Now you can randomly select any value from i_val & j_val arrays.
Note: Make sure that your pixel values will be 1 or 0. If your values are 0 and 255 then change this command
positive = data[data['value'].isin([255])]
1 Comment
vishnu saini
Thank you..! It leads to lot of calculation. i tried np.argwhere(img) to get coordinates and random.shuffle() to get random coordinates but i got TypeError: 'NoneType' object is not subscriptable in sampling only 100 random coordinates.
Use below code:
import numpy as np
y_idx, x_idx = np.where(image==1) #list of all the indices with pixel value 1
for i in range(0,100):
rand_idx = np.random.choice(x_idx) #randomly choose any element in the x_idx list
x = x_idx[rand_idx]
y = y_idx[rand_idx]
#--further code with x,y--
Comments
here is another answer that might save some memory storage, or at least a for loop..
import numpy as np
random_image = np.random.uniform(0, 1, size=(2000, 2000)) > 0.5
sel_index = np.random.choice(np.argwhere(random_image.ravel()).ravel(), size=100)
random_x, random_y = np.unravel_index(sel_index, random_bin.shape)
I just like the usage of unravel :)
PIL--> Python Imaging Library module might help you.