Outfit recognition

There are two components for that solution: mobile application and browser widget. Mobile app allows to take a new picture or select the existing one, while widget scans for the images on the browser page. Several shape adaptation algorithms are the foundation of this solution.

First algorithm recognize clothing category on the image: trousers, a dress, a skirt, etc. Every category has its own probability value to match. Categories with the maximum value are used for selection of similar objects. Second algorithm classifies every pixel of uploaded image and divides them into two groups: person and background. Hence, we get the outline of a clothed person. Now we are able to determine the length of sleeve and skirt, see if there is a collar and find out other clothing characteristics that are relative to the human body. Third algorithm converts clothing data into set of numbers which correspond to the color, pattern, etc. These values are used to select similar items of clothing: the algorithm searches for the most similar values among the clothing database.

Every image from the retailer’s catalogs is processed by the algorithm and also receives a set of numbers to generate such database.