Survey On Evaluation Methods Of Image Segmentation Algorithms

Image segmentation is an important processing step in many image, video and computer vision applications. Extensive research has been done in creating many di?erent approaches and algorithms for image segmentation, but it is still di?cult to assess whether one algorithm has better performance than another. It often needs trial and error. To overcome this issue, image segmentation evaluation techniques needs to be studied. To date, the most common method for evaluating the e?ectiveness of a segmentation method is subjective evaluation, in which a human visually compares the image segmentation results for separate segmentation algorithms, which is an unstable tedious process. Objective evaluations can to some degree avoid these problems , thus have great values of research.In this paper we first conducted a survey on the existing evaluation methods. Advantages and shortcomings of the underlying design mechanisms in these methods are discussed and analyzed through analytical and empirical examinations. Some efficient and generic supervised evaluation measures based on precision and recall index are then proposed, their features are examined on a database of 200 nature images with corresponding human segmentations. Then 4 different color image segmentation algorithms and 2 SAR image segmentation algorithms are separately evaluated using those proposed measures by comparing the segmented images with ground truth. All the experiments show that the proposed measures can precisely evaluate segmentation algorithms, provide guidance on choosing best algorithms or best parameters for a given application, and have the ability of real-time processing and general treatments.