Thread by @zahiruddin_t

How Canva efficiently removes duplicates from billions of images

T𝘄𝗶𝗻 𝗧𝗿𝗼𝘂𝗯𝗹𝗲𝘀: 𝗛𝗼𝘄 𝗖𝗮𝗻𝘃𝗮 𝗖𝗼𝗻𝗾𝘂𝗲𝗿𝘀 𝗕𝗶𝗹𝗹𝗶𝗼𝗻𝘀 𝗼𝗳 𝗜𝗺𝗮𝗴𝗲𝘀 𝘄𝗶𝘁𝗵 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆
🎯 #Refactor75 ➡Day41 #systemdesign #content #engineers

Canva has more than 20 billion user-uploaded images, with 50 million uploaded daily.

Dealing with a large variety of media creates difficulties when it comes to moderating and minimizing unnecessary duplicate content

To solve this problem, Canva uses perceptual hashing with an internally built reverse image search system

Perceptual hashing, also known as visual hashing or image fingerprinting, is a technique used to create a compact digital representation of an image or video frame based on its visual content.

The goal of perceptual hashing is to generate a hash value that is unique to the image and remains relatively unchanged even if the image undergoes minor transformations such as compression, color correction, and brightness

To perform a reverse image search using perceptual hashing, we can compare two hash values by calculating the Hamming distance between them.

Hamming distance is a metric used to measure, compare and evaluate the similarity or dissimilarity between two binary strings

Benefits for Canva

1. Reduced Storage Costs
2. Near-unique images served on lookup
3. Content Moderation: Takedown of known illegal images within seconds

Please follow @zahiruddin_t for more such content