direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Logo der TU Berlin

Details zu Abschlussarbeiten

Zur Übersicht


Clothing Description from Images to be Used in a Smart Online Shopping Application


Bachelor Thesis

The project aims at developing an application that finds and retrieves available clothing being sold online to the user matching the style of the user. The most important part of the application will be the smart clothing detection module which detects the clothing on an uploaded photo or from the image captured by a camera (e.g. shirt, trousers, bag etc.) and describe each of these clothing w.r.t their colors, textures and outlines autonomously.

Thesis Work Description: This part of the project will only focus on further labeling the clothing on the images already classified based on their types (e.g. shirt, trousers etc.). These specific labels (e.g. long, ekose, striped, blue) will describe the cloth and help the application do the text-searching through the online-shopping sites. The descriptive labels, i.e. extracted features, are currently grouped as color, texture and outline descriptives [1]. The tasks to develop this module are listed below:
1. Finding or constructing a database of clothes already grouped with respect to their types (e.g. shirts, trousers, hats, dresses etc.)
2. Feature extraction of each image in the database based on color, texture and outline [1].
3. Clustering the entire database based on the descriptive features extracted. Note that, the clustering here is descriptive clustering of the clothing. That is, there will be different clusters under each of three descriptives, i.e. color (e.g. blue shades, green shades etc.), texture (e.g. striped) and outline (e.g. long, short, sleeved etc.). → Python / C / C++ is expected for OpenCV.
4. Labeling meaningfully each of the clusters defined in step 3. As exemplified in step 3, it is expected that the labels here will be specific terms generically used in the fashion community, so that the text searching will return the similar clothing as expected. The labels could be tested in Google-Shopping.


[1] E. Hsu, C. Paz, and S. Shen, “Clothing Image Retrieval for Smarter Shopping,” EE368 - Department of Electrical and Electronics Engineering, Stanford University, 2011.


Projekt Datei:  BA Clothing.pdf

Diplomarbeit Datei:  


Betreuer:  Orhan Can Görür
Email:   orhan-can.goeruergt-arc.com


Noch Offen