A Review of Image Segmentation Evaluation in the 21st Century

In image engineering, computer vision and image pattern recognition, image segmentation plays an important role. It consists of subdividing an image into its constituent parts and extracting those parts of interest (objects). In the past 50 years, many research works have been conducted in this area, a large number of image (and video) segmentation techniques have been proposed and utilized in various applications. With many algorithms developed, some efforts have been spent also on their evaluation, a review for the efforts in the last century can be found in (Zhang, 2001). The first comprehensive review on image segmentation evaluation has been made nearly 20 years ago (Zhang, 1996). The existing evaluation methods for segmentation algorithms have been classified into analytical methods and empirical methods. The analysis methods treat the algorithms for segmentation directly by examining the principle of algorithms while the empirical methods judge the segmented image to indirectly assess the performance of algorithms. Furthermore, the empirical methods can be still classified into empirical goodness methods and empirical discrepancy methods. The empirical goodness methods judge the segmentation results according to some predefined (goodness) criteria while the empirical discrepancy methods determine the quality of segmented images by comparing to some reference images. Empirical evaluation is practically more effective and usable than analysis evaluation (Zhang, 1996). Recent advancements for segmentation evaluation are mainly made by the development of empirical evaluation techniques. In this article, after providing a list of evaluation criteria and methods proposed in the last century as background, a review of the research works made in this century (till now) for empirical evaluation of image segmentation will be provided. These new techniques are classified, comparing to the last century-developed techniques, into 3 groups: those based on existing techniques, those made with modifications of existing techniques, and those used dissimilar ideas than that of existing techniques. A comparison of these evaluation methods is made before going to the future trends and conclusion.