The Computer Vision Laboratory was established in 2003 to undertake research work in the areas of computer vision and pattern recognition. The lab received over Rs 10 million in external funding since 2003. Current research projects include development of fast image registration approaches, non-rigid structure from motion, object recognition and analysis of satellite imagery for rural applications.
Structure from motion approaches have been very successful in recovering 3D structure of a scene from a video captured by a moving camera. We are currently investigating extraction of structure of highly non-rigid objects such as articulated human motion, periodic motion or multiple independently moving rigid objects from a video sequence taken by a single moving camera.
Project Team: Ijaz Akhtar, Sohaib Khan, in collaboration with Dr Yaser Sheikh, CMU
Traditional video codecs use motion compensation based on Sum of Absolute Differences (SAD) match metric. We have shown that using a linear intensity change model works much better for compression of videos containing brightness and contrast changes. Moreover, the optimal match metric for use with such a model is the correlation coefficients.
Project Team: Arif Mahmood, Sohaib Khan
Related Publications:
Arif Mahmood, Sohaib Khan, "Early Termination Algorithms for Correlation Coefficient Based Block Matching", International Conference on Image Processing, ICIP, San Antonio, Texas, Sept 16-19, 2007
Arif Mahmood, Zartash Afzal Uzmi, Sohaib Khan, "Video Coding with Linear Compensation", in International Conference on Communications, ICC, Glasgow, June 24-28, 2007
Aerial photography can be a low cost alternate to satellite imagery for developing countries. However, for aerial photographs to be useful for GIS applications, they need to have geo-coordinates. We are developing fast registration approaches based on normalized cross correlation for registration of aerial photographs to archival satellite imagery without using expensive on-board telemetry sensors.
Project Team: Arif Mahmood, Sohaib Khan
Related Publications:
Arif Mahmood, Sohaib Khan, "Exploiting Inter-Frame Correlation for Fast Video to Reference Image Alignment", in ACCV '07, Tokyo, Japan, November 2007
Sohaib Ahmad Khan, PhD (Faculty)
Arif Mahmood (PhD Candidate)
Ijaz Akhtar (PhD Student)
Aamer Zaheer (PhD Student)
Syed Farooq Ali (MS Student)
Sohail Sadaqat (Research Associate)
Kashif Murtaza (Research Associate)
Haroon Idrees (Research Associate)
