AlchemyAPI this week exposed a computer vision API to extend its text analytics cloud service in a way that will make it easier for publishers to identify images that are closely related to one another. One of the reasons that publishers spend so much money on creating content is that right now it’s challenging to manage a library of images effectively. A publisher may have tagged a particular image with a keyword, but it’s difficult to surface the right image using limited search engine technologies.
For example, AlchemyAPI CEO Elliot Turner says that a search for the term “striped cat” in an image library is likely to return not only every image of a cat but also tigers and any other item that has the word cat included in its tag. In contrast, the new AlchemyVision service is designed to enable users of the Alchemy cloud service to see an image of a striped cat and then return every image that actually looks similar to that image to the service via the computer vision API.
The computer vision platform supports a number of image analysis functions that developers can invoke. The first function being made available is an Image Tagging API that eliminates the need to manually tag image libraries. Turner says it is designed to enable the processing of millions of photos every hour. Best of all, Turner says, the system requires no human intervention. When provided with an image file or URL, AlchemyVision returns up to 20 keywords, summarizing scenes, objects and stylistic features. A complementary Image Extraction API will perform a scan of the page, find the most prominent image and retrieve the associated image file directly.
In addition, Turner says, the AlchemyVision algorithms continue to learn the more they are invoked and have the ability to identify thousands of high-level visual concepts, including complex composite ideas such as “man wearing a hat and sunglasses, holding a fish on a lake.” Turner says the computer vision API has applications that go well beyond publishing. It could, for example, be used in life sciences applications to identify types of organisms or be deployed in conjunction with a surveillance system.
While AlchemyVision doesn’t work with live streaming video, organizations can take snapshot images of video every second and automatically feed them into AlchemyVision to identify what is in a particular image in a matter of minutes. Rather than hiring people to attach tags to images and other types of content, Turner says the Alchemy cloud service is designed to automate the tagging of content using a text analytics engine. The tagging of content has been the bane of content management systems ever since the web became a publishing vehicle. It’s high time some artificial intelligence was applied to a process that has been far too manually intensive to be as effective as it should be.