Remove black borders on images with watermarks in Python
border, image, opencv, python
Solution
To remove black borders:- Follow this link(Perfect Answer I think) :- Crop black edges with OpenCV To remove black border by specifying region, follow this link How to crop an image in OpenCV using Python Instead of cropping any part from image, you may take only ROI (Region of Interest). To do this, follow this link, How to copy a image region using opencv in python?
To remove watermark:- If watermark may appear anywhere in your image means, you cannot clear watermark fully. Just you may apply blurring effect on that image. It will blur your watermark. Its link : https://opencv-python-tutroals.readthedocs.org/en/latest/py_tutorials/py_imgproc/py_filtering/py_filtering.html If watermark will exist only on the black border means, the above mentioned methods will solve your problem.
Problem
I have a bunch of image I would like to uniformise by removing black borders. Usually I use the Trim function of Imagemagick with the fuzz parameters but in the case the image have some watermark the result is not here. Actually I'm making some tests with opencv and morphological transform to try to identify watermark and image and then select the bigger element but I'm really new with opencv and I struggle. Watermark can be everywhere, from bottom left to upper right. I would prefer a Python code but using some app like Imagemagick or similar is welcome. Actually using opencv only I get this result: ``` import copy import cv2 from matplotlib import pyplot as plt IMG_IN = '/data/black_borders/island.jpg' # keep a copy of original image original = cv2.imread(IMG_IN) # Read the image, convert it into grayscale, and make in binary image for threshold value of 1. img = cv2.imread(IMG_IN,0) # use binary threshold, all pixel that are beyond 3 are made white _, thresh_original = cv2.threshold(img, 3, 255, cv2.THRESH_BINARY) # Now find contours in it. thresh = copy.copy(thresh_original) contours, hierarchy = cv2.findContours(thresh,cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE) # get contours with highest height lst_contours = [] for cnt in contours: ctr = cv2.boundingRect(cnt) lst_contours.append(ctr) x,y,w,h = sorted(lst_contours, key=lambda coef: coef[3])[-1] # draw contours ctr = copy.copy(original) cv2.rectangle(ctr, (x,y),(x+w,y+h),(0,255,0),2) # display results with matplotlib # original original = original[:,:,::-1] # flip color for maptolib display plt.subplot(221), plt.imshow(original) plt.title('Original Image'), plt.xticks([]),plt.yticks([]) # Threshold plt.subplot(222), plt.imshow(thresh_original, cmap='gray') plt.title('threshold binary'), plt.xticks([]),plt.yticks([]) # selected area for future crop ctr = ctr[:,:,::-1] # flip color for maptolib display plt.subplot(223), plt.imshow(ctr) plt.title('Selected area'), plt.xticks([]),plt.yticks([]) plt.show() ``` results: