Bilkent University
Department of Computer Engineering


Region Recognition and Information Analysis on Multimedia Documents Using Invariant Features


Utku Can Yücel
MSc Student
Computer Engineering Department
Bilkent University

Categorization on video documents and image collections is a time consuming process. Obtaining information from specific frames on a large video stream or fetching a specific image document from a large collection by manual scanning requires too much human effort. In order to automatize these steps, it is possible to analyse multimedia documents and extract relational features. There exist several state of art methodologies in the literature like SIFT (Scale Invariant Feature Transform) and HOG (Histogram of Oriented Gradients). These methodologies can be applicable to analyse image content and establish links between given samples and queries. Our approach is to come up with a complete solution which can categorize multimedia files by analyzing information obtained by recognized regions. As an initial consideration of our research, we interested in face detection & recognition in personal photo albums to practice image processing and machine learning methodologies. As a further step, we are currently trying to handle image region matching problem to categorize the context of multimedia files. Throughout the research steps, we mainly focused on invariant features to analyse different image regions which include logo, face and text fields with different rotation, scale and position qualities.


DATE: 04 May, 2015, Monday @ 16:15