ai-agents-ocr-preprocessing-text-detection-cropped-regions-have-excessive-whit-108ed0b0

THEN Tighten the crop around the detected text to leave only 1-2 pixels of whitespace. Use OpenCV to find the bounding rectangle of the text contours (e.g., cv2.boundingRect) and crop accordingly. Ensure the cropped image margins match the training data's whitespace distribution.

import cv2
# Assume 'img' is the detection crop
# Convert to grayscale, threshold, find contours
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
_, thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
x, y, w, h = cv2.boundingRect(np.vstack(contours))
# Add small margin (1-2px)
margin = 2
tight_img = img[y-margin:y+h+margin, x-margin:x+w+margin]