Fusion of Measures for Image Segmentation Evaluation

Image segmentation is defined as the partition of an image into homogenous and meaningful constituent parts. These parts are referred to as segments. Image segmentation serves as a prerequisite step in object detection and computer vision systems. Hence, the quality of the image segmentation results has a direct impact on those systems. A lot of attention has been rendered to the development of image segmentation algorithms. However, evaluation of these algorithms has received far less attention. As a consequence, there is still no universally acceptedmeasure for comparing the performance of various segmentation algorithms, or even different parameterizations of the same algorithm. In order to develop or improve segmentation algorithms, it is crucial to evaluate the quality of their results. In Zhang et al. [1], the importance of having application-independent methods for comparing results produced by different segmentation algorithms or different parameterizations of the same algorithm is stated in terms of: The need for autonomous selection from among possible segmentations yielded by the same segmentation algorithm, the need to place a new or existing segmentation algorithm on a solid experimental and scientific ground and the need to monitor segmentations results in real time.