Product development /

Image similarity available on the Peltarion platform

January 28 2021/6 min read

Have you been in a situation where you need to search for images that are similar to a specific design but you find it hard to describe with words and harder to search for it? Or would you like to offer your customers an easier way to visually find items in a product inventory by skipping hard-coded keywords? As of today, you are able to build image similarity search models on the Peltarion platform!

From a business perspective, let’s take a look at what similarity is, what it can do for you and some examples of the opportunities this capability offers.

02/ What is image similarity?

Image similarity gives you a quantifiable measure of how alike two or more images are compared to each other. The criteria used to determine what is similar and what is not, is very subjective and can be adjusted by you depending on the use case the image similarity application will be used. For instance, you can look for images that share the same color palette, have the same patterns and forms or display products alike to the one in the input image.

Search and find items that share a pattern

03/ Where can I apply image similarity?

Image similarity relies deeply on the content of the images as opposed to the metadata description of each image. Therefore, visual similarity search overcomes the drawbacks of traditional text-based search engines that are applied to visual representations as graphics, images, videos, etc. You can apply image similarity in any scenario where the content of your query search is better described by an image. Here are some examples:

  • Enhance an existing search engine by adding visual search in spare part inventories or product catalogs
  • Display alternative products in case of items are out-of-stock or sold out
  • Locate duplicates entries on digital assets and bank of images for removal or deduplication
  • Identify the best-matching products, models or designs from a repository of referenced images
  • Track proprietary visual assets, product pictures or stock photography on websites even if the assets have been altered or if they appear in derivative works.
  • And more...

Let’s deep dive into a specific industry as e-commerce where the use of images to describe products is widely extensive and how image similarity brings tangible opportunities.

Image similarity in e-commerce

If you are running a retail store you can offer your customers the possibility to search by image for items they have seen online or in real life; for example, a piece of furniture, a design object or a painting in an office. Your potential customer can upload a picture of the item to run an image search and in return, they would get the closest matching results from your product catalog.

The benefit for your customer? A much faster and more seamless shopping experience allowing: 

  • Improve product discovery and customer engagement
  • Increase sales by reducing path to purchase and improve conversions 
  • Enhance customer experience by removing the need to guess keywords that represent the item they are looking for or unnecessary browsing through multiple menus or items they’re not interested in.

After all, a picture is worth a thousand words.

04/ Why a solution on the Peltarion platform?

Building your image similarity model on the Peltarion platform doesn’t require you to write code. This is one of the greatest advantages that Peltarion provides in comparison with some offerings in the market. In addition:

  • Despite the no-code nature of the platform, you are able to select, customize and experiment with the models to be used behind your solution
  • Your model will work for exactly your specific needs as it’s trained and fine-tuned on your proprietary data and for your requirements 
  • Easily update your model(s) when needed and the one-click deployment directly from the UI accelerates operationalization
  • The platform manages and maintains the required infrastructure and supports the full end-to-end deep learning model development with reliable security and scalability. Check here all the amazing platform features.

05/ Image classification vs image similarity?

A common question you might be asking yourself when trying to apply AI to your projects is what is the difference between image similarity and image classification. Image classification aims to predict a label or a predefined category for a specific image while image similarity quantifies the degree of similarity between two or more images.

For example, given a picture of a bird, an image classification model would predict the type of bird on the image based on a predefined set of bird labels e.g. Crested Auklet or Indigo Bunting. On the other hand, the image similarity model will return images of the most similar birds to the input picture in the given dataset. You can build models to perform both image classification and image similarity tasks on the Peltarion platform.

An image classification model predicting the type of bird from the input image

An image similarity model identifying the most similar images to the input picture

06/ Try it yourself!

Get a feel for the power of image similarity search by building your own visual search engine for your business. For some inspiration, follow this tutorial that uses image similarity to find out what fruit you (probably) got. Are you interested in deep dive into the technical side and get an idea of how image similarity actually works, this is a great post by my colleague Romain: Image similarity with deep learning explained.

If you want to explore more in detail how to build your own similarity model for your project or need some guidance, please get in touch with me at We’d love to hear from you!

  • Liliana Lindberg

    Liliana Lindberg

    Solutions Architect

    Liliana works as a solutions architect at Peltarion guiding customers to solve business challenges by using AI. She is passionate about emerging technologies and before joining Peltarion she worked for a number of years at Google as a GCP customer engineer. Her academic background includes BSc. Systems and Computing Engineering; MSc. in Geographical Information Systems from the University of Calgary, Canada; and a Master’s level Business Leadership Specialization from Duke University.

02/ More on Product development