With millions of photos floating around the Internet, it seems impossible to determine what could be popular and what might not. Well, a researcher from the Massachusetts Institute of Technology may have found a way to find out. Aditya Khosla, a Ph.D. student at MIT’s Computer Science and Artificial Lab (CSAIL), has developed an algorithm that attempts to accurately predict how popular an image is going to be.
The extensive research was based on information gathered from over two million Flickr images via automatic analyses of image content (such as colors, textures, and gradients) and social cues (how many tags the photo has, the total photo count of the user, number of user contacts, and the average views for every image of the user).
The algorithm considers other factors, like as the presence or absence of particular objects. According to Khosla, images containing the following objects received the most praise: miniskirt, maillot (a type of swimsuit), bikini, cup, brassiere, perfume, and revolver. On the other end of the spectrum, the least popular objects include: spatula, plunger, laptop, golfcart, and space heater.
Khosla is the same person behind the algorithm designed to tweak actors’ headshots to make them more memorable. On his site, the student provides a tool that allows you to upload an image and quickly see its popularity on a scale of 1 to 10. Khosla hopes to build on the algorithm and provide a similar uploader that will automatically edit your image to make it as popular as it can be.
- Google’s Squoosh will get an image web-ready with in-browser compression
- Google Lens’ Images feature recognizes objects you might want to buy
- To be or not to be 280 characters: All of Shakespeare’s works in a single tweet
- Images of Google Home Hub leak ahead of company’s October hardware event
- Did a leaked image reveal Apple’s upcoming desktop monitor?