Detect image similarity between two images in Python – Machine Learning

Do you know how Facebook can detect faces in your pictures or Google find similar pictures? Today I am going to show you one powerful method for image comparison by artificial intelligence with deep learning. Do not be concerned if you are a tech geek. I’ll write this code out for you with as little complexity as possible but will make it still do all the heavy lifting behind the scenes.

What Are We Building?

We are making a program that can look at two images and tell us how similar they are. Think of it like having a smart assistant that can tell you, “These two photos are almost identical” or “These pictures are completely different.

How Does It Work?

We’re using something called VGG16, which is a digital brain that has already learned to understand images. It‘s been trained by looking at millions of pictures, so it’s really good at understanding what’s in an image. Imagine it as a super-experienced art critic who can break down any picture into its essential features.

The Process: Step by Step

1. Image Preparation

First, we preprocess the images so the AI can recognize them. A bit like adjusting a photo’s size to get it to fit in a picture frame, we resize all our images to 224 x 224 pixels and fix their colors differently.

from keras.applications.vgg16 import VGG16
model = VGG16(weights='imagenet', include_top=False)

def extract_features(img_path, model):
img = load_img(img_path, target_size=(224, 224))
img_data = img_to_array(img)

2.Feature Extraction

The AI looks at each image and breaks it down into what we call “features” – these are like the DNA of the image. It might notice things like edges, colors, shapes, and patterns.

3. Similarity Check

We compare these features using something called “cosine similarity” – think of it as measuring the angle between two arrows. If the arrows point in the same direction, the images are similar. If they point in opposite directions, the images are different.

def calculate_similarity(img_path1, img_path2, model):
features1 = extract_features(img_path1, model)
features2 = extract_features(img_path2, model)
similarity = cosine_similarity([features1], [features2])
return similarity[0][0]

4. Understanding the Results

The program gives us a score between 0 and 1:
– A score above 0.99 means the images are practically identical
– A score above 0.5 suggests the images have some things in common
– A low score means the images are quite different

if similarity_score > 0.5:
print("The images are somewhat similar.")
elif similarity_score > 0.99:
print("The images are very similar.")
else:
print("The images are not similar.")

5.Real-World Applications

This technology has many practical uses:
- Finding duplicate photos in your collection
- Detecting copyright violations
- Searching for similar products in online shopping
- Content moderation on social media
- Medical image analysis
-------------------------------------------------------------------
For those interested in the technical details:
- We're using the VGG16 architecture, which is a convolutional neural network (CNN)
- The model has been pre-trained on ImageNet, a massive database of labeled images
- We remove the top classification layers (include_top=False) because we only want the feature extraction capabilities
- The cosine similarity metric helps us compare high-dimensional feature vectors

Feel free to try this code with your own images! Just remember to install the required libraries (keras, numpy, scikit-learn) and have your image files ready in the same directory as your code.

*Note: This code requires TensorFlow/Keras and scikit-learn to be installed on your system.*