Creating an inverse image search index with deep learning

Because the whole point of an autoencoder is to learn an encoding or a low-dimensional representation of a set of images, they make for great feature extractors. Furthermore, we can use them as the perfect building blocks of image search indices, as we'll discover in this recipe.

Getting ready

Let's install OpenCV with pip. We'll use it to visualize the outputs of our autoencoder, in order to visually assess the effectiveness of the image search index:

$> pip install opencv-python

We'll start implementing the recipe in the next section.

How to do it…

Follow these steps to create your own image search index:

1. Import the necessary libraries:

   import cv2
   import numpy as np
   from tensorflow.keras import Model
   from tensorflow.keras.datasets import fashion_mnist
   from tensorflow.keras.layers import *

2. Define build_autoencoder(), which instantiates the autoencoder. First, let&apos...