2010 - 20XX
2008 - 2010
2003 - 2008

Teknikringen 14, 5th floor, room 512 (show on map)
Tel:
Email: sobhannp@kth.se

 Sobhan Naderi Parizi

  "As of Sep-2010 I am PhD student at CS department of University of Chicago"

  Formerly, master student in
Systems, Control, & Robotics
A joint program between
Department of Computer Science and
Department of Electrical Engineering

 
Royal Institute of Technology
(KTH)
Stockholm, Sweden


Education:

M.Sc. in Systems, Control, & Robotics
Royal Institute of Technology (KTH), Stockholm, Sweden
Sep 2008 to Sep 2010
Thesis title: Reducing Ambiguity of Local Descriptors for Visual Recognition (PDF)

B.Sc. in Computer Engineering (software discipline)
Amirkabir University of Technology (Tehran Polytechnic), Tehran, Iran
Sep 2003 to Jan 2008
Thesis title: Adaptive continuous Hidden Markov Models integrated with language grammars. Link to thesis page.


Research Interests:
The main focus of my recent works has been on Computer Vision and particularly the following problems:
  • Visual object detection/recognition
  • Contextual priming in image classification and action recognition
  • Spatial-aware BoFs (Bag of Features)
  • Semantic image segmentation
  • Content-based image classification
  • Kernel-based learning methods (SVM, MKL)
  • Probabilistic learning methods
I have also been working on Automatic Speech Recognition (ASR) problem during my bachelor's final project where I implemented a general toolbox for training and manipulating Hidden Markov Models (HMMs).

Publications:

M. M. Ullah, S. Naderi Parizi, I. Laptev
Improving Bag-of-Features Action Recognition with Non-local Cues (PDF)
British Machine Vision Conference (BMVC'10), Aberystwyth, UK, Sep. 2010

S. Naderi Parizi, I. Laptev, A. Tavakoli Targhi
Modeling Image Context using Object Centered Grids (PDF) (PPT)
Int. Conf. on Digital Image Computing: Techniques and Applications (DICTA'09), Melbourne, Australia, Dec. 2009

S. Naderi Parizi, A. Tavakoli Targhi, O. Aghazadeh, J.O. Eklundh
Reading Street Signs Using a Generic Structured Object Detection and Signature Recognition Approach (PDF) (PPT)
Int. Conf. on Vision Application (VISAPP'09), Lisbon, Portugal, Feb. 2009

A.S. Shahmiri, S. Naderi Parizi, M.K. Akbari
A New Error Correction Code (PDF) (PPT)
Int. Conf. on Latest Advances in Networks (ICLAN'07), Paris, France, Dec. 2007

Journal Papers:

A. Askary, A. Masoudi-Nejad, A. Mizbani, R. Sharafi, S. Naderi Parizi, M. Purmasjedi
N4: A Precise and Highly Sensitive Promoter Predictor Using Neural Network Fed by Nearest Neighbors (PDF)
Journal of Genes and Genetic Systems, Nov. 2009


Teaching:

Teacher Assistant In "Compiler Design Principles" Course, Amirkabir University of Technology, Spring 2007
Teacher Assistant In "Analysis and Design of Algorithms" Course, Amirkabir University of Technology, Spring 2007
Teacher Assistant In "Data Structures And Algorithms" Course, Amirkabir University of Technology, Fall 2007


Projects:
Action Recognition in Realistic Scenarios:
We extract motion features of actions as well as appearance features of scenes and make an empirical evaluation on contribution of each feature channel on recognition of human actions in realistic scenarios. Evaluations are done on HOHA2 dataset. We have also annotated the dataset by bounding boxes of actors.

Image Classification using Semantically Decoupled BoFs:
Images are decomposed into three semantically homogineous segments. The segments are determined based on substantial similarity of image regions using a texture-transform method. This type of image decomposition helps to disambiguate contextual discriminability of BoF histograms.
     

Spatial-Aware BoFs using Object Centered Grids (OCG):
Location information is incorporated into BoF framework by formation of spatial grids w.r.t. salient object(s). We show consistent improvement over the baseline approach (fixed spatial grids) on the image classification challenge of PASCAL'07 database.
Fixed-Grid
 
OCG

Urban Place Recognition:
This project is done as a part of MOBVIS research project. In this project we propose an approach to detect street plates in an image and to automatically recognize street-names. Having a searchable map installed on your mobile phone, this project provides your mobile phone with the functionality to automatically find your exact geo-location on the map by only taking a photo of a street plate around you.

Speech Recognition (LISSH project):
A full package of useful tools for continuous speech recognition tasks (incl. continuous HMM). To read more about the project and download executable files click here.

Personal:

Attended Conferences

Hobbies