Tuesday 20th of February 2018

A Comparative Study about Object Classification based on Global and Local Features

Hammad Naeem, Maria Minhas and Jameel Ahmed

Scene classification and object recognition is a hot area of research in the field of computer vision and has always fascinated researchers to explore strategies for optimization of results. Global and local features are manipulated to find a match in the images or scene categories.This paper mainly comprises of finding the scene labels based on the objects present in it.The image is transformed into a feature space and the classifier is trained to differentiate each class in the feature space.Various feature extraction techniques like RGB histogram , SIFT and co-variance are explored in this paper to find an optimized result. Different classifiers were tested individually as well as their combinations to achieve better results. Combination of Sparse SIFT and Dense SIFT techniques was found to perform better compared to others.

Keywords: Scene Classification, object recognition, Bag ofwords, Sparse SIFT, Dense SIFT.

Download Full-Text


Hammad Naeem
Hammad Naeem received his degree of Master in Computer Vision and Robotics from Heriot Watt University-Edinburgh, University of Girona-Spain and University of Burgundy-France. He was on an Erasmus Mundus program which was funded by European Union. He holds Bachelors in Electrical Engineering from Air University, Islamabad. Currently he is serving as a Lecturer in Electrical Department at HITEC University of Engineering and Technology, Taxila Pakistan. He has particular interests in Vision based robotics and image processing based intelligent machines.

Maria Minhas
Maria Minhas is currently doing his Masterís in Electrical Engineering from University of Minnesota, Duluth, USA. She did his Bachelor in Electrical Engineering from HITEC University, Taxila, Pakistan. She has research interest in image processing and Bio-medical vision based solutions.

Jameel Ahmed
HITEC University, Taxila Pakistan

IJCSI Published Papers Indexed By:





IJCSI is a refereed open access international journal for scientific papers dealing in all areas of computer science research...

Learn more »
Join Us

Read the most frequently asked questions about IJCSI.

Frequently Asked Questions (FAQs) »
Get in touch

Phone: +230 911 5482
Email: info@ijcsi.org

More contact details »