An Evaluation Measure of Image Segmentation Based on Object Centres
Classification of organic materials obtained from rock and drill cuttings involves finding multiple objects in the image. This task is usually approached by segmentation. The quality of segmentation is evaluated by matching the whole detected objects to a reference segmentation. We are interested in representing each object by a single reference point called the “centre”. This paper proposes an evaluation measure of image segmentation for such representation. We argue that measures based only on distance between obtained centres and a set of predefined centres are insufficient. The proposed measure is based on a list of desirable properties of the segmentation. The three components of the measure evaluate the under/over segmentation of the objects, the proportion of centres placed in the background rather than in objects, and the distance between the guessed and the true centres. The ability of the measure to distinguish between segmentation results of different quality is illustrated on three sets of examples including an image containing microfossils and pieces of inert material.
KeywordsImage segmentation evaluation measures discrepancy methods microfossils palynomorphs
Unable to display preview. Download preview PDF.
- 8.Elisseeff, A.B.H.A., Guyon, I.: A stability based method for discovering structure in clustered data. In: Proc. Pacific Symposium on Biocomputing, pp. 6–17 (2002)Google Scholar
- 11.Charles, J.J., Kuncheva, L.I., Wells, B., Lim, I.S.: Object location within microscopic images of palynofacies (submitted, 2006)Google Scholar
- 13.Wang, Y., Chou, J.: Automatic segmentation of touching rice kernels with an active contour model. Transactions of the ASAE 47(5), 1803–1811 (2004)Google Scholar
- 14.Paul, G.V., Beach, G.J., Cohen, C.J.: A realtime object tracking system using a color camera. In: Applied Imagery Pattern Recognition Workshop, pp. 137–142 (2001)Google Scholar