How can the location of the paper rectangle be detected as a rotated in these images using opencv? The big problem is the background which makes it hard to extract only the paper.

I manage to get parts of the paper edges with hough lines but there are also some false lines for example.

Here are the images where I need the paper to be detected. As you can see, the camera position stays the same.

• Example image 1
• Example image 2
• Example image 3
• Example image 4
• Example image 5

I’ve tried these things:

• Simple Threshholding -> Result: Since parts of the background have the same color as the paper, i cant get the outline of the paper
• Hough Lines -> Result: Depending on how you change the parameters, too many false lines are detected or too view lines of the actual paper edges
• Contour detection and Shape Approximation -> Result: A proper threshholding would be needed, to work with contours
• Feature Matching -> Result: Simply not enough good features to match
• Shifting a rect over the images -> Takes way to long and is not precise, since the rect is a little distorted, and not perfectly square

The best results I got was a thresholding with otsu, then an edge detection with canny, extracting contours and then houghLinesP:

import cv2         
import numpy as np         
img = cv2.imread("path/to/image", cv2.IMREAD_GRAYSCALE)         
_, binary = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)         
contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
edges = cv2.Canny(contours, 50, 150) 
lines = cv2.HoughLinesP(edges, 1, np.pi / 180, 100, minLineLength=45, maxLineGap=8)