Use python opencv dilation with a mask-CodePudding

From what I have seen in the documentation, I can't use opencv dilate function with a mask.

Any idea how to achieve this?

I mean, I would like to do a dilate operation only within a ROI on an image.

Thank you very much in advance.

CodePudding user response：

Follow the following steps to get your results:

Extract your ROI from the image. This will give you a new image of only the ROI.
Perform your dilation operation on the extracted image of the ROI.
Replace the ROI region of the original image with this dilated image.

Following can be the possible code. Here the image is Img, and ROI is given my rectangle [x, y, w, h].

ROI_Img = Img[y:y h, x:x w].copy()
Kernel = np.ones((3, 3), dtype=uint8)
ROI_Img_Dilated = cv2.dilate(ROI_Img, Kernel, iterations=1)
Img[y:y h, x:x w] = ROI_Img_Dilated.copy()

CodePudding user response：

As of now, there are no functions available to perform selective morphological operations.

I have outlined a naïve approach below along with the code:

Input image:

Code:

img = cv2.imread('hand_drawn_contours.jpg',1)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
# binarize the image
th = cv2.threshold(gray,0,255,cv2.THRESH_BINARY cv2.THRESH_OTSU)[1]

# finding contours
contours, hierarchy = cv2.findContours(th, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

We will perform dilation only for the first contour present in contours. Consider the following to be our ROI:

# create a blank mask
mask = np.zeros((img.shape[0], img.shape[1]), np.uint8)
# draw selected ROI to form the mask
mask = cv2.drawContours(mask,contours,1,255, -1)

When performing morphology the shape of the ROI either increases or decreases. To maintain the same shape; the same morphology operation must be performed on the mask also:

# dilate both the image and the mask
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7,7))
dilate_th = cv2.dilate(th, kernel, iterations=1)
dilate_mask = cv2.dilate(mask, kernel, iterations=1)

# using the dilated mask, retain the dilated ROI
im1 = cv2.bitwise_and(dilate_th, dilate_th, mask = dilate_mask)

Now to preserve the other contours as is, invert the mask to retain them

dilate_mask_inv = cv2.bitwise_not(dilate_mask)
im2 = cv2.bitwise_and(th, th, mask = dilate_mask_inv)

res = cv2.add(im1, im2)