FuncAI PHP

# Image similarity

Detect how similar two given images are

# Demo

Image 1
Click
to select your own image
Image 2
Click
to select your own image

Given two images, this application calculates how similar these images are.

# Use it for

  • Showing similar images to your users
  • Add an "image search" feature to your app, where users can upload an image and you find matching images in your database
  • Grouping / filtering similar images
  • Detecting different sizes of the same image

# Installation

After you've followed the installation instructions, run the following to install the application:

php vendor/funcai/funcai-php/install-bitmr50x1-featurevector.php

# Usage

$model = new \FuncAI\Models\BitMR50x1();
$result = $model->predict([
  $pathToInputImage
]);

The result will be a so called "feature vector". A feature vector is an array of float values. You can either predict the feature vectors of two or more images at once and compare them, or you can store the feature vector in your database. To calculate the similarity, you need to use an algorithm like the "cosine similarity". See our example file (opens new window) for a sample implementation.

There are also other more complex, but more performant ways to find similar feature vectors if you have a large amount of feature vectors. Interesting projects to get started with large scale feature vector comparisons are faiss (opens new window) and milvus (opens new window).

Also, if you are using Elasticsearch, it has native support (opens new window) to calculate the cosine similarity.

← More →