A study for selecting a metric for a first level evaluation of image segmentation methods

Image segmentation is one of the first important parts of the image analysis and understanding. Theo Pavlidis initially introduced it in late 60s, and since then researchers have studied it and produced a great variety of algorithms. These algorithms provide segmentation results that could be characterized from “objective”, like Pavlidis segmentation to “subjective” ones dependent on the interpretation of the human user. In this paper we attempt to initiate a first stage evaluation of color image segmentation methodologies by proposing a simple segmentation metric. In particular, we have used three segmentation algorithms on different types of images (human faces/bodies, natural environments, and structures (buildings)). Then we apply the metric (formula) on these segmented results in order to quickly observe and evaluate the performance (behavior) of each segmentation methodology.