/plushcap/analysis/algolia/algolia-engineering-how-we-handled-color-identification

How to handle color identification in search | Algolia

What's this blog post about?

Searching by color on ecommerce websites can be challenging due to limitations in text-based search indexes. To improve the accuracy of color searches, a system that automatically identifies the color of objects in thumbnails is needed. This article explores an approach similar to Vue.ai's method for identifying foreground colors in queryable words. The process involves several steps such as cropping images, shrinking them down to 100x100px, separating background from the main object, removing skin tones, isolating color clusters, and categorizing cluster average colors into readable English names using a K-Nearest-Neighbors algorithm. The experiment is available on GitHub as an open-source Python script that can be used to enrich search indexes with extracted color tags.

Company
Algolia

Date published
Aug. 8, 2023

Author(s)
Léo Ercolanelli

Word count
1472

Language
English

Hacker News points
None found.


By Matt Makai. 2021-2024.