[Python] OpenCV rectangle detection accuracy on picture

I'm a junior developer, and for my project, I need to detect rectangles on images, specifically thumbnails from the Netflix dashboard. I'm discovering OpenCV, which seems like the best option to achieve what I want, but I'm not getting good results. Some thumbnails are not being detected, and I don't understand why. My code looks like this, and I've tested many preprocessing modifications on my image, but I got the best results with these parameters:

import cv2
import numpy as np

def nothing(x):
pass

img = cv2.imread("./photos/photo.png")

while True:
img_copy = img.copy()

hsv = cv2.cvtColor(img_copy, cv2.COLOR_BGR2HSV)

lower_bound = np.array([1, 0, 0])
upper_bound = np.array([180, 255, 255])

mask = cv2.inRange(hsv, lower_bound, upper_bound)
kernel = np.ones((5,5), np.uint8)
erosion = cv2.erode(mask, kernel)

contours, _ = cv2.findContours(erosion, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

for contour in contours:
epsilon = 0.05 * cv2.arcLength(contour, True)
approx = cv2.approxPolyDP(contour, epsilon, True)

if len(approx) == 4:
x, y, w, h = cv2.boundingRect(contour)
if w >= 100 and h >= 100:
# Dessiner un rectangle autour du contour
cv2.rectangle(img_copy, (x, y), (x + w, y + h), (0, 255, 0), 2)

cv2.imshow("Image with Contours", img_copy)

if cv2.waitKey(1) & 0xFF == ord('q'):
break

cv2.destroyAllWindows()


I got this result:



The result before contour tracing (erosion) looks like this:



What confuses me even more is that I don't understand why it’s specifically the undetected images that aren't being detected.

We have tried various types of preprocessing techniques: Grayscale Conversion, Thresholding, Canny Edge.

Continue reading...