28 Oct 2015 EveryPixel – Intelligent Image Search from The Pressfoto Group

Pressfoto Group, the company behind Pressmaster, Dragon Images, and Pressfoto stock agency, launched their new intelligent image search product EveryPixel today.

Currently in open beta, EveryPixel is a cross-agency image search platform using the fast-growing technology known as neural networking. The new service was inspired by some of Pressfoto’s previous services presents a new approach to image search for licensable photos, making it easier to find and license stock photos by searching for specific ideas or related images.

Background

Pressfoto Group is led by its co-owner, microstock celebrity Dmitriy Shironosov. In late 2013, the company released two different products. MediaSTAMP is aimed at content creators as a copyright protection service that embeds copyright information in image files with invisible and trackable codes. Microstock.photos targeted photo buyers as an image meta-search engine that could search for images from multiple agencies and compare them for price and minimum purchase requirement, directing buyers to the original sources for purchases.

Soon, they decided to merge these initiatives into a product that would serve both sides of the market in one integrated platform: EveryPixel.

The Technology

EveryPixel applies artificial intelligence to image analysis. Using neural networking, a variety of machine learning, Pressfoto expects to change the way people search for images online.

Normally, finding licensable images involves three steps. A buyer starts with an idea for an image or a concept they want to use, converts that idea into possible keywords, then searches for those keywords. EveryPixel skips the keywords, by jumping straight from a sample image to results. On their website, customers can upload an image, sourced from anywhere or snapped by themselves, and the search engine will automatically analyze the content to deliver results.

While reverse image search is not new technology, and giants like Google offer it, Dmitriy says that the deep learning in EveryPixel’s neural networks sets them apart from those systems. He also says this technology “can do magic”: it finds subjects through mathematical analysis, automatically grouping images and building topic trees. This could be a great keywording tool for microstock contributors; although due to the constant learning performed on algorithms these keywords would have a useful life of about 2 years – their analytics show that images earn 90% of their revenue in the first 2 years.

The Service

Although EveryPixel intends to be an open interface for all copyrighted image hubs, they’re initially focusing the services on microstock agencies, since they are companies well versed in image copyright.

The main purpose of the product is to connect content creators and potential buyers. Users can get exact match results, uploading an image they found online but don’t know where to legally source it. They are then directed to the channel(s) where they can buy the image, while at the same time being presented with a selection of similar alternatives. It intends to offer better exposure of stock photos for contributors, adding intelligence to search results and keywording.

There’s an interesting secondary feature that allows customers to get similar images results, but based on concept similarity rather than visual similarity. The learning algorithms can analyze the meaning in a photo provided and find professional images that may be visually different but have the same subject or meaning.  This works as a sort of “visual query,” where a buyer creates a brief using photos instead of words. In the future, they want to add a sketching function, for buyers to simply submit drawings of their ideas and get image results.

All results are displayed on a price comparison board that shows all available sizes for images and converts credit prices into USD. To purchase, buyers are directed to the original source.

The platform collects image data either via APIs or using web crawlers. To date, 140 million images are indexed, sourced from 20 different agencies.

EveryPixel just went public in beta to test their performance in the market. If the project is successful, they will soon add contributor functionalities such as market analysis and an image copyright protection system. And they already have some plans for the future as well: adding a feature to analyze sketches, as mentioned above, and the introduction of a distribution platform using agency APIs, enabling contributors to upload once and automatically distribute to multiple channels.