Image Processing with SciPy and NumPy in Python
Image processing is used in areas like computer vision and medical imaging, focusing on enhancing and analyzing digital images. In Python, NumPy treats images as arrays for efficient pixel-level operations, while SciPy’s ndimage module provides tools for filtering and transformations, enabling fast and lightweight processing.
Installation
Ensure you have the required libraries installed:
pip install numpy scipy matplotlib imageio
Opening and Writing Images
To begin any image processing task, the first step is to load and visualize the image. We'll use imageio.v3 to read an image and matplotlib to display it.
Example:
import imageio.v3 as iio
import matplotlib.pyplot as plt
img = iio.imread(r'C:\Users\visha\OneDrive\Desktop\Python\racoon.png')
plt.imshow(img)
plt.axis('off')
plt.title("Curious Raccoon")
plt.show()
Output
Explanation: iio.imread() loads the image into a NumPy array. plt.imshow() visualizes it and plt.axis('off') hides axes for a cleaner look.
Creating NumPy array from the image
An image is essentially a multi-dimensional NumPy array. Knowing its shape and data type is important for applying filters and transformations.
import imageio.v3 as iio
import numpy as np
img = iio.imread('raccoon.png')
print("Shape:", img.shape)
print("Data type:", img.dtype)
Output
Explanation: Shape helps understand the image layout (e.g., 266x341x3 for RGB). Data type (usually uint8) shows pixel value range (0–255).
Creating RAW file
A .raw file stores raw binary data from an image sensor or matrix. It's useful when dealing with uncompressed data in image pipelines.
Example: Creating RAW file using SciPy
import imageio.v3 as iio
import numpy as np
img = iio.imread('raccoon.png')
img.tofile("raccoon.raw")
Output
Explanation: tofile() saves the image pixel data as a binary file, useful for low-level image processing.
Opening RAW File
To work with .raw files, we use np.fromfile() to reconstruct the image data into a usable NumPy array.
import numpy as np
img = np.fromfile('raccoon.raw', dtype=np.uint8)
print(img.shape)
Output
Explanation: fromfile() reads binary data and the array must be reshaped manually if you want to visualize it (e.g., reshape to original height × width × channels).
Getting Statistical Information
Understanding the min, max and average pixel intensity gives insight into brightness, contrast and histogram distribution of the image.
import numpy as np
img = iio.imread('raccoon.png')
print("Max:", img.max())
print("Min:", img.min())
print("Mean:", img.mean())
Output
Explanation: Max and min values indicate contrast and Mean gives an overall idea of brightness.
Cropping the Image
Cropping helps focus on a particular region of interest (ROI) in an image by slicing the NumPy array.
import imageio.v3 as iio
import matplotlib.pyplot as plt
img = iio.imread('raccoon.png')
x, y, _ = img.shape
# Crop center region
crop = img[x//3: -x//8, y//3: -y//8]
plt.imshow(crop)
plt.axis('off')
plt.title("Cropped Raccoon")
plt.show()
Output
Explanation: img.shape gives image dimensions (height x, width y, channels _). img[x//3: -x//8, y//3: -y//8] selects a central region using slicing and plt.imshow() visualizes the cropped section.
Flipping Image (Vertical)
Flipping an image (up-down or left-right) is a common data augmentation technique in image preprocessing.
import imageio.v3 as iio
import matplotlib.pyplot as plt
import numpy as np
img = iio.imread('raccoon.png')
flipped = np.flipud(img)
plt.imshow(flipped)
plt.axis('off')
plt.title("Flipped Image (Up-Down)")
plt.show()
Output
Explanation: np.flipud() flips the image along the vertical axis.
Filtering images
Filtering is a fundamental technique in image processing used to enhance or suppress certain features. It helps in tasks like smoothing, sharpening and edge detection.
1. Gaussian Blur
Blurring helps reduce image noise and details using a Gaussian kernel. It’s useful in preprocessing steps like edge detection or thresholding.
from scipy.ndimage import gaussian_filter
import matplotlib.pyplot as plt
img = iio.imread('raccoon.png')
blurred = gaussian_filter(img, sigma=5)
plt.imshow(blurred.astype(np.uint8))
plt.axis('off')
plt.title("Gaussian Blurred")
plt.show()
Output
Explanation: gaussian_filter(img, sigma=5) smooths the image using a Gaussian kernel. sigma controls the intensity of blur and converts to uint8 before display to ensure proper color rendering.
2. Sharpening Image (Unsharp Masking)
Sharpening increases contrast between edges to enhance details and clarity. Unsharp masking subtracts a blurred version from the original.
from skimage.color import rgb2gray, rgba2rgb
from scipy.ndimage import gaussian_filter
import imageio.v3 as iio
import matplotlib.pyplot as plt
img = iio.imread('raccoon.png')
if img.shape[-1] == 4:
img = rgba2rgb(img)
gray = rgb2gray(img).astype(float)
blur = gaussian_filter(gray, 5)
alpha = 30
sharp = gray + alpha * (gray - gaussian_filter(blur, 1))
plt.imshow(sharp, cmap='gray')
plt.axis('off')
plt.title("Sharpened Image")
plt.show()
Output
Explanation: Converts image to grayscale using rgb2gray. gray - gaussian_filter(blur, 1) extracts edge details and adds edge details back using alpha scaling Unsharp Masking.
Denoising Images
Image denoising removes random noise to enhance image quality, particularly useful in low-light photography or scanned documents.
Setup & Imports
import numpy as np
import matplotlib.pyplot as plt
from scipy.ndimage import gaussian_filter, median_filter, rotate, sobel
from skimage.color import rgb2gray, rgba2rgb
import imageio.v3 as iio
1. Add noise
Artificial noise is added to simulate a noisy environment, commonly seen in real-world low-light or sensor-imperfect images.
img = iio.imread('raccoon.png')
if img.shape[-1] == 4:
img = rgba2rgb(img)
gray = rgb2gray(img).astype(float)
noise_img = gray + 0.9 * gray.std() * np.random.random(gray.shape)
plt.imshow(noise_img, cmap='gray')
plt.axis('off')
plt.title("Noisy Image")
plt.show()
Output
Explanation: Adds random values scaled by image standard deviation to simulate real-world noise (e.g., from low-light sensors).
2. Gaussian Denoising
Gaussian filtering smooths the image by averaging pixel values with its neighbors using a Gaussian kernel, effectively reducing high-frequency noise.
denoised = gaussian_filter(noise_img, sigma=2.2)
plt.imshow(denoised, cmap='gray')
plt.axis('off')
plt.title("Denoised (Gaussian)")
plt.show()
Explanation: Smooths the image using a Gaussian kernel to reduce high-frequency noise while preserving structure.
Edge Detection using Sobel Filter
Sobel edge detection identifies image edges by computing intensity gradients using 3×3 kernels. It highlights boundaries by combining horizontal and vertical changes, aiding in tasks like segmentation and object detection.
import numpy as np
import matplotlib.pyplot as plt
from scipy.ndimage import rotate, gaussian_filter, sobel
im = np.zeros((300, 300))
im[64:-64, 64:-64] = 1
im = rotate(im, 30, mode='constant')
im = gaussian_filter(im, sigma=7)
plt.imshow(im, cmap='gray')
plt.axis('off')
plt.title("Original Synthetic Image")
plt.show()
dx = sobel(im, axis=0, mode='constant')
dy = sobel(im, axis=1, mode='constant')
sobel_edges = np.hypot(dx, dy)
plt.imshow(sobel_edges, cmap='gray')
plt.axis('off')
plt.title("Sobel Edge Detection")
plt.show()
Output
Explanation: Creates a synthetic image, applies Gaussian blur, then detects edges using Sobel filters by computing horizontal and vertical gradients and combining them to highlight edge intensity.
