Has an “I love that! Where can I get it?” moment ever happened to you? You are reading a magazine or walking down the street, as you notice someone’s outfit. Maybe it’s a t-shirt with a beautiful motif and you want to have something similar. Visual search can help to get what you want.
With visual search and object recognition, consumers now can simply snap an image and search for a specific item of interest. There is no need anymore to write a long list of keywords into a search box, which is sometimes hard to do (it’s much easier to imagine it than to describe). It is a fun and fast way of searching.
The flow goes like this: you see something you like, snap an image and upload it to a webshop. The system will recognize fashion items in the image and highlight them. Then you select one of the highlighted areas and choose whether to search by category OR/AND color. Products that look similar to the selected one will be displayed first.
By analyzing uploaded images, a brand can get an idea of styles, looks and trends their customers are interested in. This can lead to picture profiling, where the system could learn a customer’s style preferences and show more relevant search results. Also, individualized campaign management would be possible.
PCM agents could also benefit from visual search. A new product could be added by uploading an image, which would be analyzed by ML. The system could check if the image is visually similar to another and suggest the creation of a variant of the existing product. It could also extract information from an image, such as category or color and populate data fields. It would speed up creation process and the chance of creating duplicated products would be less.
How is the solution different?
According to this blog post, there are some common problems with visual search for e-commerce. Here are the issues we address with our solution:
- Incorrect product type – With item recognition, product types are highlighted and a customer can select one to search for visually similar products within.
- Search was slow – Solr is used for search, so search results are displayed fast.
- Color match did not work – Color is one of the features extracted from the image and fed into the search.
- Lack of omnichannel integration – Search results are in the same format as for standard text search. A customer can choose to add the product to the cart or pick up in store.
- Search algorithm wrong – Since LIRE is used, the solution is extensible and different image descriptors can be used to suit your needs.
If you like the article, please rate it and leave a comment. Stay tuned for a technical post on how visual search can be implemented and integrated into SAP Hybris Commerce.