An evaluation metric for image segmentation of multiple objects

It is important to be able to evaluate the performance of image segmentation algorithms objectively. In this paper, we define a new error measure which quantifies the performance of an image segmentation algorithm for identifying multiple objects in an image. This error measure is based on object-by-object comparisons of a segmented image and a ground-truth (reference) image. It takes into account the size, shape, and position of each object. Compared to existing error measures, our proposed error measure works at the object level, and is sensitive to both over-segmentation and under-segmentation. Hence, it can serve as a useful tool for comparing image segmentation algorithms and for tuning the parameters of a segmentation algorithm.