Semiautomatic Segmentation with Compact Shape Prior

Piali Das, Olga Veksler, Vyacheslav Zavadsky, Yuri Boykov

In Canadian Conference on Computer and Robot Vision (CRV), pp.28-36, June 2006.

Abstract

We present a semiautomatic segmentation algorithm, that can segment an object of interest from its background based on a single user selected seed. We are able to obtain reliable and robust segmentation with such low user interaction by assuming that the object to be segmented is of "compact" shape. We base our work on the powerful Graph Cut segmentation algorithm of Boykov and Jolly [2]. An additional benefit of incorporating the compact shape prior we are able to bias the graph cuts segmentation framework towards larger objects. It helps to counteract the well known bias of [2] to shorter segmentation boundaries. Segmentation results are quite sensitive to the choice of parameters, and so another contribution of our paper is that we show how to select the parameters automatically. We demonstrate the effectiveness of our method on the challanging industrial application of transistor gate segmentation in an integrated chip, for which it produces highly accurate results in real-time.


WHOLE PAPER: pdf file (2,641 Kb)