Finding shadows in images using Python's scikit-image package

Shadows. They can make photos look dramatic and cool, or they can be a total pain for image editing. 😎 But detecting them? That’s actually a super useful skill for understanding the geometry and lighting of a scene.

In this post, I’m going to show you how to hunt down shadows in an image using Python and scikit-image. We’ll dive into why the HSV (Hue, Saturation, Value) color space is the secret sauce for this task and build a simple pipeline to catch those shadows. 🕵️

By the end of this, you’ll know:

Color Spaces: Why RGB isn’t always your best friend.
Thresholding: How to let the computer decide what’s “dark” and what’s not.
Morphological Operations: Cleaning up the mess (noise) in your masks.

Prerequisites

Grab these libraries if you haven’t already:

# %pip install numpy scikit-image matplotlib

import numpy as np
import skimage
import matplotlib.pyplot as plt
from skimage import filters, morphology, color, io, data

Step 1: Loading a Sample Image

Let’s use the classic “stereo motorcycle” image from scikit-image. It’s got some nice, distinct shadows for us to play with.

image_rgb = data.stereo_motorcycle()[0]

# Display the original image
plt.figure(figsize=(8, 6))
plt.imshow(image_rgb)
plt.title("Original Image")
plt.axis("off")
plt.show()

Step 2: Why HSV?

Most images are stored in RGB (Red, Green, Blue). But here’s the thing: in RGB, a shadow makes everything darker—red, green, and blue values all drop. This makes it tough to tell a dark object apart from a shadow.

Enter HSV (Hue, Saturation, Value)! ✨

Hue: The actual color (red, blue, etc.).
Saturation: How “intense” the color is.
Value (Brightness): How light or dark it is.

Shadows mostly mess with the Value (making it darker) but leave the Hue alone. This is our superpower! We can isolate shadows just by looking for dark areas that still have the right color properties.

# Convert the image to HSV color space
image_hsv = color.rgb2hsv(image_rgb)

# Extract Hue and Value (Brightness) channels
hue = image_hsv[:, :, 0]
brightness = image_hsv[:, :, 2]

Step 3: Thresholding

Now we need to decide: “How dark is a shadow?” Instead of guessing, we’ll use Otsu’s method. It’s a smart algorithm that automatically finds the best threshold to separate the foreground from the background.

We’ll apply it to both brightness and hue. Shadows are usually:

Darker than the rest (Low Brightness).
Have a specific hue profile (we’ll use a threshold to help separate them).

# Find global threshold on the lightness and hue to detect dark (shadow) regions
brightness_thresh = filters.threshold_otsu(brightness)
hue_thresh = filters.threshold_otsu(hue)

print(f"Brightness Threshold: {brightness_thresh:.2f}")
print(f"Hue Threshold: {hue_thresh:.2f}")

Step 4: Creating the Shadow Mask

Time to make the mask! We’ll mark pixels as “shadow” if they are darker than our brightness threshold AND meet our hue criteria.

# Create a mask for the shadow regions
# We look for pixels that are darker than the threshold AND have a lower hue value
shadow_mask = (brightness < brightness_thresh) & (hue < hue_thresh)

# Remove small noise from the mask using morphological opening
# This removes small isolated spots that are likely not shadows
shadow_mask = morphology.remove_small_objects(shadow_mask, min_size=500)

Step 5: Visualizing the Result

Let’s see what we caught! 👀 We’ll overlay our mask on the original image.

# Display the original image and the shadow mask
fig, ax = plt.subplots(1, 2, figsize=(16, 8))

# Original Image
ax[0].imshow(image_rgb)
ax[0].set_title("Original Image")
ax[0].axis("off")

# Shadow Mask
# We invert the mask for visualization if needed, or just show the mask itself.
# Here, we'll show the mask where yellow indicates the shadow.
ax[1].imshow(shadow_mask, cmap="viridis")
ax[1].set_title("Detected Shadows (Yellow)")
ax[1].axis("off")

plt.tight_layout()
plt.show()

Conclusion

And boom! 💥 By switching to HSV, we easily separated the shadows from the rest of the image. Otsu’s method did the heavy lifting for the thresholds, and a little cleanup made our mask look sharp.

This isn’t just a cool party trick; it’s a fundamental step for things like shadow removal, scene understanding, and object tracking. Now go forth and bust some shadows! 👻