An Efficient Method for Optimizing Segmentation Parameters

An Efficient Method for Optimizing Segmentation Parameters

Jacob D'Avy (University of Tennessee, USA), Wei-Wen Hsu (Old Dominion University, USA), Chung-Hao Chen (Old Dominion University, USA), Andreas F. Koschan (The University of Tennessee, USA) and Mongi Abidi (University of Tennessee, USA)
DOI: 10.4018/978-1-4666-9685-3.ch002
OnDemand PDF Download:


Segmenting an image into meaningful regions is an important step in many computer vision applications such as facial recognition, target tracking and medical image analysis. Because image segmentation is an ill-posed problem, parameters are needed to constrain the solution to one that is suitable for a given application. For a user, setting parameter values is often unintuitive. We present a method for automating segmentation parameter selection using an efficient search method to optimize a segmentation objective function. Efficiency is improved by utilizing prior knowledge about the relationship between a segmentation parameter and the objective function terms. An adaptive sampling of the search space is created which focuses on areas that are more likely to contain a minimum. When compared to parameter optimization approaches based on genetic algorithm, Tabu search, and multi-locus hill climbing the proposed method was able to achieve equivalent optimization results with an average of 25% fewer objective function evaluations.
Chapter Preview

1. Introduction

Image segmentation is the process of dividing an image into regions of interest. It is a crucial step in image analysis applications such as facial recognition (Ferrera, Franco & Maio, 2012), target tracking (Heber, Godec, Rüther, Roth & Bischof, 2013), and medical imaging (Chen, Udupa, Bagci, Zhuge & Yao, 2012). Segmentation is a difficult problem which has no general solution. Free parameters are often required to constrain a segmentation problem to produce the desired result. However, the relationship between parameter values and segmentation output is not always clearly defined and finding the best segmentation parameters for an image can be a tedious process. The motivation for parameter optimization is to simplify this process by allowing segmentation methods to work in varying conditions with limited user interaction. There are several challenges associated with optimization in this setting that are described in Crevier (2008).

  • 1.

    There is no differentiable function to relate segmentation performance with parameters.

  • 2.

    There are multiple local minima in the parameter space.

  • 3.

    The process of segmenting and evaluating is time expensive.

These challenges restrict the type of optimization that can be applied to segmentation parameters. The most basic approach to finding optimal parameters is a brute-force search of the possible parameter values (Ilea & Whelan, 2009). This method is straightforward but requires evaluating a very large set of parameter combinations which makes it computationally infeasible. Other approaches have used neural networks to find optimal parameters for segmenting images (Shen, Sandham, Granat & Sterr, 2005). Parameter values were incorporated into network weights and optimized using a set of training data. These approaches are limited to applications where training data is available and a universal segmentation goal can be defined. The most common approach to segmentation parameter optimization has been to use a direct search in the parameter space using methods such as local searches and stochastic searches which do not require differentiation. The typical search process moves through a solution space by making small changes to a candidate solution as it seeks to minimize an objective function. The advantage to direct search methods is that they are relatively efficient in finding local minima in the parameter space and they are also easily adaptable to different segmentation applications.

Complete Chapter List

Search this Book: