Comparing Clothing Styles by Means of Computer Vision Methods

摘要

The paper deals with a problem of comparing and retrieving visual data representing various clothing styles. Proposed solution joins several sophisticated computer vision methods, such as face detection using AdaBoost strategy, human body segmentation and decomposition using pictorial structures and appearance models, together with visual descriptors employing simplified dominant color descriptor. The input images do not necessary have to be taken in controlled environment, so the flexibility of the system is high. The proposed algorithm makes it possible to compare images presenting humans and retrieve images with similar clothing style. The potential application is the area of social network services, mostly recommendation web-based systems, that help people choose clothes and share with clothing ideas. Developed algorithm has been tested on 650 images gathered from various social media in the Internet and showed high accuracy rate.