Spatio-temporal wardrobe generation of actors’ clothing in video content

Abstract

In this paper, we propose a methodology for spatio-temporal wardrobe generation for video content. The main goal is to suggest relevant matches between clothes worn by actors and images originating from a set of e-commerce clothing sites. The semi-automatic generation of fine-grained spatial metadata for each video sequence is based on shot detection, keyframe detection, feature matching and clothing type classification based filtering. The result of this annotation process is a spatio-temporal database consisting of videos and the corresponding actor clothing. This database can be queried in various ways depending on the intended target application.