Implementation of watershed segmentation algorithm in OpenCV

Key takeaways:

  • The watershed algorithm segments overlapping and connected objects in images effectively.

  • It overcomes limitations of traditional methods like thresholding.

  • OpenCV is an open-source library for computer vision and machine learning.

  • Implementation involves preprocessing and applying the watershed algorithm.

  • Preprocessing includes image loading, grayscale conversion, and noise reduction.

  • The algorithm identifies connected components and marks regions for separation.

  • It enhances segmentation precision beyond traditional techniques.

The watershed algorithm of OpenCV is used for classic image segmentation. The images having objects that are overlapping and connected by boundaries can be easily segmented by using a watershed algorithm. Traditional algorithms like thresholding and contour detection sometimes cannot detect the objects in the images. But by using the watershed algorithm, we can efficiently perform image segmentation.

Sample images that we can use for the watershed algorithm are given below:

Sample image
Sample image

OpenCV

OpenCV is an open-source library of computer vision and machine learning. This library includes all the computer vision and machine learning methods, like image segmentation, object detection, face recognition, and many other machine learning methods. To include this library, run the following command in the terminal:

!pip install opencv-python

Let’s dive into the implementation

A watershed algorithm is an OpenCV function used for image segmentation. We divide the code into two steps. The first step is related to the preprocessing required for the image segmentation, and the second step contains the main implementation of the watershed algorithm.

Step 1: Preprocessing

Here is the code for preprocessing:

import cv2 
import numpy as np 
from matplotlib import pyplot as plt

img = cv2.imread("checkerboard.png") 
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

ret, binary_img = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU) 
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3)) 
binary_img = cv2.morphologyEx(binary_img, cv2.MORPH_OPEN, kernel,iterations=2) 

fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(5, 5))

sure_bg = cv2.dilate(binary_img, kernel, iterations=3)
axes[0, 0].imshow(sure_bg, cmap='gray')
axes[0, 0].axis('off')
axes[0, 0].set_title('Sure Background')

dist = cv2.distanceTransform(binary_img, cv2.DIST_L2, 5)
axes[0, 1].imshow(dist, cmap='gray')
axes[0, 1].axis('off')
axes[0, 1].set_title('Distance Transform')

ret, sure_fg = cv2.threshold(dist, 0.5 * dist.max(), 255, cv2.THRESH_BINARY)
sure_fg = sure_fg.astype(np.uint8)
axes[1, 0].imshow(sure_fg, cmap='gray')
axes[1, 0].axis('off')
axes[1, 0].set_title('Sure Foreground')

unknown = cv2.subtract(sure_bg, sure_fg)
axes[1, 1].imshow(unknown, cmap='gray')
axes[1, 1].axis('off')
axes[1, 1].set_title('Unknown')
Preprocessing for watershed algorithm

Code explanation

Here’s a line-by-line breakdown of the code:

  • Lines 1–3: We import the required libraries. OpenCV (cv2) is used for image processing, NumPy (np) for numerical operations, and matplotlib.pyplot (plt) for visualizing the images.

  • Lines 5–6: The image is loaded using cv2.imread and then converted to grayscale using cv2.cvtColor.

  • Lines 8–10: We apply the cv2.threshold to the gray image. Then, we set the kernel by using cv2.getStructuringElement. To remove the noise, we applied the cv2.morphologyEx, which follows two steps: erosion followed by dilation. This step helps to remove the noise and smooth the contour of a large object in the binary image.

  • Line 12: We use plt.subplots, which creates the figure and arranges the subplots in two rows and two columns.

  • Lines 14–17: We use cv2.dilate on the binary_img to increase the image’s bright regions. Then save the results in the sure_bg. Then, plot the results with the title of sure background.

  • Lines 19–22: Then we use the cv2.distanceTransform on the binary_img, which measures the distance of each white pixel to the nearest dark pixel in the binary_img. Then plot with the name of Distance Transform.

  • Lines 24–28: We use cv2.threshold with a threshold value 0.5 on dist variable to obtain the foreground in sure_fg variable. Then, plot with the title of the sure foreground.

  • Lines 30–33: To get the unknown areas in the image, we use cv2.subtract on sure_bg and sure_fg. Then, we plot the unknown areas with the title unknown.

Step 2: Implementation of the watershed algorithm

Here is the code of the implementation:

ret, markers = cv2.connectedComponents(sure_fg)
markers += 1
markers[unknown == 255] = 0

fig, axes = plt.subplots(nrows=1, ncols=3, figsize=(10,10))
axes[0].imshow(markers, cmap="tab20b")
axes[0].axis('off')
axes[0].set_title('Markers')

# Watershed Algorithm
markers = cv2.watershed(img, markers)
axes[1].imshow(markers, cmap="tab20b")
axes[1].axis('off')
axes[1].set_title('“Watershed Result”')

labels = np.unique(markers)
obj_arr = []
for label in labels[2:]:
    target = np.where(markers == label, 255, 0).astype(np.uint8)
    contours, hierarchy = cv2.findContours(target, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    obj_arr.append(contours[0])

img_with_contours = img.copy()
img_with_contours = cv2.drawContours(img_with_contours, obj_arr, -1, color=(0, 23, 223), thickness=2)

axes[2].imshow(cv2.cvtColor(img_with_contours, cv2.COLOR_BGR2RGB))
axes[2].axis('off')
axes[2].set_title('Image with Contours')
Implementation of watershed algorithm

Code explanation

Here’s a line-by-line breakdown of the code:

  • Lines 1–3: In these lines, we use the cv2.connectedComponents to identify the connected objects in the sure_fg. Then, we add one to markers, so that the background is one instead of zero. Then, mark the region of unknown with zero.

  • Lines 5–8: We use plt.subplots, which creates the figure and arranges the subplots in one row and three columns. Then, plot the marker by setting the cmap="tab20b".

  • Lines 11–14: The cv2.watershed function takes two parameters, img and markers. Then, we plot the results with the title Watershed Result.

  • Lines 16–21: To obtain the contour objects in the whole image, we are using a loop to iterate over the unique markers named as label starting from 2, because we ignore the unknown and background part of the image. In this loop, we use the cv2.findContours function that obtains the contour in the image and then appends it to obj_arr.

  • Lines 23–28: We use cv2.drawContours to outline the original image named img. Then we plot the results with the title, “Image with Contours.”

Conclusion

In conclusion, the watershed algorithm is an effective image segmentation technique for separating overlapping objects, and it can be easily implemented using OpenCV in Python. Its ability to refine segmentation surpasses traditional methods like thresholding and contour detection.

Frequently asked questions

Haven’t found what you were looking for? Contact Us

What is the principle of watershed algorithm?

The watershed algorithm is based on the concept of treating the image as a topographic surface, where pixels represent elevation. It segments the image by simulating the flooding of the surface and identifying boundaries where water would meet.

What are the advantages of the watershed algorithm?

Advantages of the watershed algorithm include its ability to effectively segment overlapping objects, refine boundaries, handle complex shapes, and provide more accurate results compared to traditional segmentation methods.

What are the applications of watershed model?

Applications of the watershed model include medical imaging (for segmenting tissues and organs), object detection in computer vision, satellite image analysis, and any scenario requiring precise delineation of regions in an image.

How does watershed work image processing?

In image processing, the watershed algorithm works by identifying local minima (representing objects) in the image and treating them as seed points for segmentation. It floods the surface from these points until boundaries between objects are reached, creating distinct segments.

Free AI Mock Interviews
Coding Interview
Coding PatternsFree Interview
Gain insights and practical experience with coding patterns through targeted MCQs and coding problems, designed to match and challenge your expertise level.
System Design
YouTubeFree Interview
Learn to design a video streaming platform like YouTube by tackling functional and non-functional requirements, core components, and high-level to detailed design challenges.

Free Resources

Copyright ©2025 Educative, Inc. All rights reserved