Mahmood Amintoosi

Research Page

This page is under construction.


Here it is a list of my favorite research area, which I had some experiments about them


Image Registration using SSIM

Here you can find the materials about the following paper:

Amintoosi, M., Fathy, M., and Mozayani, N., “Precise image registration with structural similarity error measurement applied to super-resolution,” EURASIP Journal on Applied Signal Processing, Volume 2009, Article ID 305479, 8 pages, (to be appear).




Super-resolution is the problem of obtaining a single high-resolution (HR) image given a set of low resolution (LR) images which are related by small displacements. This definition is about to multi-frame super-resolution as the most important variation of Super-resolution (SR) methods.

The left top image shows one LR image from a sequence containing 55 frames. The bottom image shows the super-resolved image from this sequence. The higher resolution of the bottom is evident. The mentioned sequence can be downloaded from Prof. Milanfar’s website, who is a known scientist in super-resolution context.

Each SR algorithm has 3 major steps: registration, reconstruction and deconvolution. Sometimes a stage named, synthesis is considered, which is usually happen in examples based SR methods.

We have the following accepted paper about dealing with moving objects in super-resolution:

·         M. Fathy, N. Mozayani, and M. Amintoosi, “Outlier removal for super-resolution problem using QR-Decomposition,” in 2008 International Conference on Image Processing, Computer Vision, and Pattern Recognition (IPCV08), USA, July 2008.



A panorama is a compact representation of the environment viewed from one 3D position. While an ordinary image can capture only a small portion of the environment, a panorama can capture it all or any portion of it, depending on the geometry in which the panoramas are represented. Ideally, the image stitching process should be completely automatic, requiring no user information in calculating the panorama. The next panorama is created by AutoStitch which is written by Dr. Matthew Brown and  Prof. David Lowe, from the following images and 3 others. (These images can be found in the demo version of AutoStitch).








Snapshot from test video


Mean and Variance


Mixture of Gaussains


The Proposed Method


Background detection is a basic step in many image processing applications. Many techniques have been proposed for this problem by researchers. Object detection is the major use of it. Object detection can be achieved by building a representation of the scene called the “background model” and then finding deviations from the model for each incoming frame. This process is referred to as the background subtraction. Among the many methods for background detection, the method of Stauffer and Grimson [2000] is very popular. They used a mixture of Gaussians to model the pixel color.

During my Computer Vision course with Dr. Farbiz, we proposed a new method for background modeling using QR-Decomposition. The basic idea behind the proposed method was finding the similar frames, via column dependency in Linear Algebra. We published three papers regarding to this idea:

·         M. Amintoosi, F. Farbiz, M. Fathy, “A QR Decomposition based Mixture Model Algorithm for Background Modeling”, ICISC2007, Sixth International Conference on Information, Communication and Signal Processing, Singapore, pp. 1-5, December 2007.

·         M. Amintoosi, F. Farbiz, M. Fathy, M. Analoui, N. Mozayani, "QR-Decomposition-based algorithm for background subtraction", ICASP2007, IEEE International Conference on Acoustics, Speech, and Signal Processing, pp. 1093-1096, April, 2007, Honolulu, Hawaii, USA.

·         M. Fathy, M. Analoui, N. Mozayani, M. Amintoosi, F. Farbiz, “A Background Model Initialization Algorithm Based on QR-Decomposition”, 4th Iranian Conference on Machine Vision and Image Processing, Mashhad, Iran, Feb, 14-15, 2007. AcrobatArm.JPG


Background Detection

Image Registration





Noise Removal with PCNN

Fish School Clustering





3D Reconstruction

Detecting Car lights





Vehicle Detection

Facial Expression Recognition





Student Sectioning

University Timetabling





Tiling Problem

N-Queen Problem & Knight’s tour in Chess




When I was a BS student, every term, I had a trouble with selecting my courses in the next term. During my C++ courses, I wrote a program, which its input was timetable of potentially courses that I could take, and its output was all of feasible timetables which I could take without any conflict.

Student Timetabling




This page was last updated on June 2008.