DEV Community

Cover image for Introduction To Computer Vision with Python (Part 1)
TheCSPandz
TheCSPandz

Posted on

Introduction To Computer Vision with Python (Part 1)

Note: In this post, we will only be working with Grey-Scale images to make it easy to follow.

What is an Image?

An image can be thought of as a matrix of values, where each value represents the intensity of a pixel. There are three main types of image formats:

  • Binary: An image in this format is represented by a single 2-D matrix with values of 0 (black) and 1 (white). It's the simplest form of image representation.

Binary Representation

  • Grey-Scale: In this format, an image is represented by a single 2-D matrix with values ranging from 0 to 255; where 0 represents black and 255 represents white. The intermediate values represent varying shades of grey.

Grey-Scale Representation

  • RGB Scale: Here, an image is represented by three 2-D matrices (one for each color channel: Red, Green, and Blue), with values ranging from 0 to 255. Each matrix contains pixel values for one color component, and combining these three channels gives us the full color image.

RGB-Scale Representation

Filters

Filters are tools used to modify images by applying certain operations. A filter is a matrix (also called a kernel) that moves across the image, performing computations on the pixel values within its window. We’ll cover two common types of filters: Mean Filters and Median Filters.

Mean Filters

A Mean Filter is used to reduce noise by averaging the pixel values within a window. It replaces the center pixel in the window with the average of all the pixel values within that window. The cv2.blur() function applies a mean filter with a kernel size of 3x3, which means it considers a 3x3 window of pixels around each pixel to compute the average. This helps in smoothing the image.

import cv2
import matplotlib.pyplot as plt

image = cv2.imread('McLaren-720S-Price-1200x675.jpg', cv2.IMREAD_GRAYSCALE)

# Applies a Mean Filter of size 3 x 3
blurred_image = cv2.blur(image, (3, 3))

plt.subplot(1, 2, 1)
plt.imshow(image, cmap='gray')
plt.title('Original Image')
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(blurred_image, cmap='gray')
plt.title('Mean Filtered Image')
plt.axis("off")
plt.show()
Enter fullscreen mode Exit fullscreen mode

Output 1

Median Filters

A Median Filter is used to reduce noise by replacing each pixel's value with the median value of all pixels in a window. It’s particularly effective in removing salt-and-pepper noise. The cv2.medianBlur() function applies a median filter with a kernel size of 3. This method replaces each pixel with the median value of the pixel values in its neighborhood, which helps in preserving edges while removing noise. Here the larger the kernel size the more blurred the image gets.

import cv2
import matplotlib.pyplot as plt

image = cv2.imread('McLaren-720S-Price-1200x675.jpg', cv2.IMREAD_GRAYSCALE)

# Applies a Median Filter with a kernel size of 3
blurred_image = cv2.medianBlur(image, 3)

plt.subplot(1, 2, 1)
plt.imshow(image, cmap='gray')
plt.title('Original Image')
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(blurred_image, cmap='gray')
plt.title('Median Filtered Image')
plt.axis("off")
plt.show()
Enter fullscreen mode Exit fullscreen mode

Output Median

Custom Filters

You can create custom filters to apply specific operations on your images. The cv2.filter2D() function allows you to apply any custom kernel to an image. The cv2.filter2D() function applies a custom kernel (filter) to the image. The kernel is a matrix that defines the operation to be performed on the pixel values. In this example, the kernel enhances certain features of the image based on the specified values.

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

image = cv2.imread('McLaren-720S-Price-1200x675.jpg', cv2.IMREAD_GRAYSCALE)

# Define a custom filter kernel
kernel = np.array([[2, -1, 5],
                   [-5, 5, -1],
                   [0, -1, 0]])

filtered_image = cv2.filter2D(image, -1, kernel)

plt.subplot(1, 2, 1)
plt.imshow(image, cmap='gray')
plt.title('Original Image')
plt.axis('off')

plt.subplot(1, 2, 2)
plt.imshow(filtered_image, cmap='gray')
plt.title('Filtered Image')
plt.axis('off')

plt.show()
Enter fullscreen mode Exit fullscreen mode

Filtered Image

Thresholding

Note: In the code snippets, you will see _ , image when assigning the thresholded image. This is because the cv2.threshold function returns two values: the threshold value used and the thresholded image. Since we only need the thresholded image, we use _ to ignore the threshold value.

Thresholding converts an image into a binary image by setting pixel values based on a condition. There are several types of thresholding techniques:

Global Thresholding

Simple Thresholding

This method sets a fixed threshold value for the entire image. Pixels with values above the threshold are set to the maximum value (255), and those below are set to 0. The cv2.threshold() function is used for simple thresholding. Pixels with intensity greater than 127 are set to white (255), and those with intensity less than or equal to 127 are set to black (0), producing a binary image.

import cv2
import matplotlib.pyplot as plt

image = cv2.imread('McLaren-720S-Price-1200x675.jpg', cv2.IMREAD_GRAYSCALE)

_, thresholded_image = cv2.threshold(image, 127, 255, cv2.THRESH_BINARY)

plt.subplot(1, 2, 1)
plt.imshow(image, cmap='gray')
plt.title('Original Image')
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(thresholded_image, cmap='gray')
plt.title('Thresholded Image')
plt.axis("off")
plt.show()
Enter fullscreen mode Exit fullscreen mode

Binary Threshold

Otsu Thresholding

Otsu's method determines the optimal threshold value automatically based on the histogram of the image. This method minimizes intra-class variance and maximizes inter-class variance. By setting the threshold value to 0 and using cv2.THRESH_OTSU, the function automatically calculates the best threshold value to separate the foreground from the background.

import cv2
import matplotlib.pyplot as plt

image = cv2.imread('McLaren-720S-Price-1200x675.jpg', cv2.IMREAD_GRAYSCALE)

_, otsu_thresholded_image = cv2.threshold(image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)

plt.subplot(1, 2, 1)
plt.imshow(image, cmap='gray')
plt.title('Original Image')
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(otsu_thresholded_image, cmap='gray')
plt.title("Otsu's Thresholded Image")
plt.axis("off")
plt.show()
Enter fullscreen mode Exit fullscreen mode

Otsu Thresholding

Adaptive Thresholding

Mean Adaptive Thresholding

In Mean Adaptive Thresholding, the threshold value for each pixel is calculated based on the average of pixel values in a local neighborhood around that pixel. This method adjusts the threshold dynamically across different regions of the image. The cv2.adaptiveThreshold() function calculates the threshold for each pixel based on the mean value of the pixel values in a local 11x11 neighborhood. A constant value of 2 is subtracted from this mean to fine-tune the threshold. This method is effective for images with varying lighting conditions.

import cv2
import matplotlib.pyplot as plt

image = cv2.imread('McLaren-720S-Price-1200x675.jpg', cv2.IMREAD_GRAYSCALE)

mean_adaptive_thresholded_image = cv2.adaptiveThreshold(image, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, 2)

plt.subplot(1, 2, 1)
plt.imshow(image, cmap='gray')
plt.title('Original Image')
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(mean_adaptive_thresholded_image, cmap='gray')
plt.title('Mean Adaptive Thresholded Image')
plt.axis("off")
plt.show()
Enter fullscreen mode Exit fullscreen mode

Mean Adaptive

Gaussian Adaptive Thresholding

Gaussian Adaptive Thresholding computes the threshold value for each pixel based on a Gaussian-weighted sum of the pixel values in a local neighborhood. This method often provides better results in cases with non-uniform illumination. In Gaussian Adaptive Thresholding, the threshold is determined by a Gaussian-weighted sum of pixel values in an 11x11 neighborhood. The constant value 2 is subtracted from this weighted mean to adjust the threshold. This method is useful for handling images with varying lighting and shadows.

import cv2
import matplotlib.pyplot as plt

image = cv2.imread('McLaren-720S-Price-1200x675.jpg', cv2.IMREAD_GRAYSCALE)

gaussian_adaptive_thresholded_image = cv2.adaptiveThreshold(image, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)

plt.subplot(1, 2, 1)
plt.imshow(image, cmap='gray')
plt.title('Original Image')
plt.axis("off")

plt.subplot(1, 2, 2)
plt.imshow(gaussian_adaptive_thresholded_image, cmap='gray')
plt.title('Gaussian Adaptive Thresholded Image')
plt.axis("off")
plt.show()
Enter fullscreen mode Exit fullscreen mode

Gaussian Adaptive

References

Top comments (1)

Collapse
 
sreno77 profile image
Scott Reno

Very interesting