The Johns Hopkins University

Whiting School of Engineering

Department of Electrical and Computer Engineering

An Interactive System for the Simultaneous Segmentation

And Recognition of Objects in Images

Seminar By

Dheeraj Singaraju

Graduate Research Assistant

Electrical and Computer

Abstract:

Object recognition is arguably one of the most important problems in computer vision and image analysis. Essentially, this problem deals with the semantic interpretation of various objects present in an image, and more specifically, with the identification and categorization of these objects. The fact that objects often appear in a cluttered scene and exhibit immense variability in their appearance across natural images, makes the problem of object recognition a challenging one.

A related problem in human and computer vision is that of object segmentation. It refers to the problem of finding within the image, the boundaries of certain objects of interest, or alternately, the regions that correspond to each of these objects. Recognition and segmentation play very interesting roles in human vision and perception. It is known that early visual areas perform a low level segmentation of the scene, but the final understanding of the scene requires a further symbiotic interplay between segmentation and recognition. While it is not clear whether one of the tasks totally precedes each other, it is apparent that this interaction helps improve the individual performance of either module.

Backed by this motivation, we propose to build an interactive system for the simultaneous segmentation and recognition of objects in images. Our system will allow the user to indicate certain objects of interest in the image, and consequently produce accurate boundaries and contextual descriptions for these objects. The problem is posed as an optimization on a discrete lattice and we show that this framework finds equivalent constructions in electrical network theory. More specifically, we use this equivalence to provide a unifying framework for the existing optimization techniques based on graph cuts and random walks on discrete lattices. In this talk, we will present preliminary work on using such equivalent electrical network constructions in order to devise general purpose object segmentation techniques.

Thursday, November 29, 2007

4:00 p.m.

Barton 117

Refreshments will be served at 3:45 p.m.

FOR DISABILITY INFORMATION

CONTACT: Candace Abel (410) 516-7031 cabel@jhu.edu