Barış Sümengen


 

Current Position: I work at Riya Inc. in San Mateo, CA

Resume (somewhat outdated): Download the PDF of my resume

Ph.D. Thesis (September 2004): Abstract
Variational Image Segmentation and Curve Evolution on Natural Images.

Consulting Link

My friend has setup a strategy and management consulting site. Some interesting stuff there such as what gadgets to buy if you are a frequent traveler.

Travel and Vacation

Are you making travel plans. Istanbul, Turkey is a good choice.
I suggest staying at Ada Hotel Istanbul (Sultanahmet area) for a smooth experience.


Software I Developed

Image segmentation software that uses Edgeflow-based Anisotropic diffusion (C++)
Both windows binary and source code are available. Based on the ideas presented on this Technical Report and some more recent stuff.

Matlab toolbox for Level Set Methods (Matlab)
This set of Matlab files implements Level Set Methods and follows Osher and Fedkiw's book. A combination of curvature-based forces, vector field-based forces and forces in  the normal direction can be used.

CIMPL Matrix Performance Library (C++)
Robust and simple (to use and understand) multi-dimensional C++ library for Matrices and Vectors. Linear algebra operations are supported mainly through BLAS and LAPACK. Useful for researchers in image processing and many other fields.

Documents

Graph Partitioning Active Contours (GPAC) for Image Segmentation
In this paper we introduce new variational segmentation cost functions that are based on pair wise similarities or dissimilarities of the pixels...

Edgeflow-driven Variational Image Segmentation: Theory and Performance Evaluation
We introduce robust variational segmentation techniques that are driven by an Edgeflow vector field...

Research Related Links

» Berkeley Segmentation Data Set
Includes a set of natural images with manual ground truth segmentations associated with them. The package include many useful Matlab source code.

» Segmentation Research at UC Berkeley

» Segmentation Research of Dr. Song-Chun Zhu at UCLA

» Netlab Library for Pattern Recognition (for Matlab)
Useful and complete toolbox. An accompanying book can also be bought.

My current research Areas:

Graph Partitioning Active Contours (GPAC)

In this project, we propose a new class of variational segmentation cost functions. Our cost functions are based on pair-wise dissimilarities between individual pixels and have been successfully applied to natural images by graph partitioning techniques. These cost functions are minimized within a variational framework. We refer to our work as graph partitioning active contours (GPAC).

Curve Evolution and Image Segmentation using Edgeflow

Starting with the Edgeflow technique, which has been shown to be highly successful on natural images, a curve evolution method is proposed. To verify the effectiveness of this technique, extensive tests are conducted on the Berkeley segmentation data set and associated ground truth. Our method is compared to Geodesic Active Contours and Gradient Vector Flow (GVF) The results show that our methods outperform the current state of the art.

Publications: Technical Report.

Anisotropic Diffusion and Image Segmentation using Edgeflow

Starting with the Edgeflow technique, which has been shown to be highly successful on natural images, an anisotropic diffusion method is proposed. Our method is compared to Perona and Malik's anisotropic diffusion and Self Snakes by conducting extensive tests on the Berkeley segmentation data set and associated ground truth. The results show that our methods significantly outperform the current state of the art.

Publications: Technical Report.

Multi-scale Edge Detection and Image Segmentation

In this project, we introduce new multi-scale techniques for edge detection and image segmentation. The idea is to favor edges that exist across multiple scales and localize them at the smallest scale they exist.

Publications: EUSIPCO 2005 paper.

Image Categorization and Content Based Image Retrieval

This project introduces a simple method for automatically collecting and categorizing images from the world wide web. Using the category structure, a search and retrieval strategy is proposed for a large database of images (around a million)

Publications: ICIP 2001 paper.

Category Pruning in Image Databases using Segmentation

Most large image databases can be divided into categories implicitly or explicitly. If this categorization is achieved by using automated techniques it is possible that there are a number of outliers in these categories. Our purpose in this project is to prune these categories using image segmentation techniques so that the precision of the category is improved with little effect to recall.

Publications: EUSIPCO 2005 paper.