Crop cellphone image with python

Question

I'm trying to crop all the images contained in a folder. I want to keep the object in the picture and make the background black by coping the cropped image and paste it to another file in the same coordinates as obtained from the source.

I have all the coordinates stored in a csv file with this structure:

Filename	xmax	xmin	ymax	ymin
a.jpg	768	4	512	8
b.jpg	1200	3	899	10

I use this code for reach my goal:

import cv2
import numpy as np
save_path = '/content/drive/MyDrive/Colab Notebooks/object_cropped'
train_path = '/content/train'

with open('min_max.csv') as csvfile:
    reader = csv.DictReader(csvfile)
    for row in reader:
      xmax = int(row['xmax'])
      xmin= int(row['xmin'])
      ymax = int(row['ymax'])
      ymin = int(row['ymin'])
 
      image = cv2.imread(os.path.join(train_path , row['Filename'])) 
      
      object_ = image[ymin:xmax, xmin:ymax] 
 
      h, w, c = image.shape
 
      black_background = np.zeros([h, w, 3])
      for y in range(h):
          for x in range(w):
              black_background[y,x] = [0,0,0] 
 
      black_background[ymin:xmax, xmin:ymax] = object_  
      cv2.imwrite(os.path.join(save_path , 'cropped_' + row['Filename']), black_background)

It works very well for images taked from the web but when i try to apply this code with pictures taked with cellphone it doesn't work.

Both images (web images and cellphone images) are jpg with horizontal and vertical resolution :96dpi bit depth: 24, the only things that are not in common is the image size.

How can i solve this problem?

EDIT: Example:

Photo:

I need to crop the object inside the boundaries box.

But when i apply the code, the result is:

Values used:

xmax = 949
xmin= 489
ymax = 1829
ymin = 181

This happens only with pictures taked with cellphone or tablet.

Answer 1

The problem is object_ = image[ymin:xmax, xmin:ymax]. The solution is:

import cv2
import numpy as np
save_path = '/content/drive/MyDrive/Colab Notebooks/object_cropped'
train_path = '/content/train'

with open('min_max.csv') as csvfile:
    reader = csv.DictReader(csvfile)
    for row in reader:
      xmax = int(row['xmax'])
      xmin= int(row['xmin'])
      ymax = int(row['ymax'])
      ymin = int(row['ymin'])
 
      image = cv2.imread(os.path.join(train_path , row['Filename'])) 
      
      object_ = image[ymin:ymax, xmin:xmax] ####the right solution
 
      h, w, c = image.shape
 
      black_background = np.zeros([h, w, 3])
      for y in range(h):
          for x in range(w):
              black_background[y,x] = [0,0,0] 
 
      black_background[ymin:ymax, xmin:xmax] = object_ ####the right solution 
      cv2.imwrite(os.path.join(save_path , 'cropped_' + row['Filename']), black_background)