Unsupervised Performance Evaluation of Image Segmentation
- PDF / 2,668,545 Bytes
- 12 Pages / 600.03 x 792 pts Page_size
- 101 Downloads / 244 Views
Unsupervised Performance Evaluation of Image Segmentation Sebastien Chabrier, Bruno Emile, Christophe Rosenberger, and Helene Laurent Laboratoire Vision et Robotique, UPRES EA 2078, ENSI de Bourges, Universit´e d’Orl´eans, 10 boulevard Lahitolle, 18020 Bourges cedex, France Received 1 March 2005; Revised 5 January 2006; Accepted 21 January 2006 We present in this paper a study of unsupervised evaluation criteria that enable the quantification of the quality of an image segmentation result. These evaluation criteria compute some statistics for each region or class in a segmentation result. Such an evaluation criterion can be useful for different applications: the comparison of segmentation results, the automatic choice of the best fitted parameters of a segmentation method for a given image, or the definition of new segmentation methods by optimization. We first present the state of art of unsupervised evaluation, and then, we compare six unsupervised evaluation criteria. For this comparative study, we use a database composed of 8400 synthetic gray-level images segmented in four different ways. Vinet’s measure (correct classification rate) is used as an objective criterion to compare the behavior of the different criteria. Finally, we present the experimental results on the segmentation evaluation of a few gray-level natural images. Copyright © 2006 Hindawi Publishing Corporation. All rights reserved.
1.
INTRODUCTION
Segmentation is an important stage in image processing since the quality of any ensuing image interpretation depends on it. Several approaches have been put forward in the literature [1, 2], . . .. The region approach for image segmentation consists in determining the regions containing neighborhood pixels that have similar properties (gray-level, texture,. . .). The contour approach detects the boundaries of these regions. We have decided to focus on the first approach, namely the region-based image segmentation, because the corresponding segmentation methods give better results in the textured case (the most difficult one). Classification methods can be used afterwards. In this case, a class can be composed of different regions of the segmentation result. However, it is difficult to evaluate the efficiency and to make an objective comparison of different segmentation methods. This more general problem has been addressed for the evaluation of a segmentation result and the results are available in the literature [3]. There are two main approaches. On the one hand, there are supervised evaluation criteria based on the computation of a dissimilarity measure between a segmentation result and a ground truth. These criteria are widely used in medical applications [4]. Baddeley’s distance [5], Vinet’s measure [6] (correct classification rate), or Hausdorff ’s measure [7] are examples of supervised evaluation criteria. For the comparison of these criteria, it is possible to use synthetic images whose ground truth is directly available. An alternative solution is to use the segmentation
results manually made by experts o
Data Loading...
