How is image similarity calculated?

Each product’s primary image and packshot get a match (100%) or no match (0%) characterization, based on our sophisticated machine learning model, Imagine. 

Our goal for image similarity is to make sure that the packaging displayed in the eRetailer’s website is the latest one you have provided us. This way the consumers’ expectations are met, as they receive in the mail the same product they saw on the eRetailer’s product page.