Performance Evaluation of Image Segmentation using Objective Methods

Background/Objectives: Image segmentation, a crucial and an essential step in image processing, determines the success of higher level of image processing. In this paper, a detailed study about different evaluation techniques based on subjective and objective methods have been discussed. Methods/Statistical analysis: An application specific characteristic of image segmentation paves a way for development of numerous algorithms. Traditionally subjective method of evaluation is used to determine the segmentation performance accuracy. As this evaluation method is quantitative and biased, a qualitative method of evaluation is demanded. This is done using the objective method of evaluation where discrepancy and goodness methods are used.Discrepancy method is used in widespread for predefined benchmark images where it has corresponding ground truth image for comparison. Goodness method is used for real time images where no ground truth image is available for comparison. These methods of objective evaluation are highly needed to validate the segmentation methods which are increasing rapidly in recent years. Findings: A detailed study of different evaluation methods are discussed and experimented over different segmentation methods. Boundary based methods like sobel, canny, susan, region based methods like region growing, thresholding and a hybrid method, combining boundary based and region based method are used for the purpose of experimentation.Experimental result shows that hybrid method performs better than other existing ones and also highlights the importance ofimage quality assessment method to identify a better segmentation technique for all type of images.