Image Quality Assessment for Image Segmentation Algorithms: Qualitative and Quantitative Analyses

Image segmentation is part of the image processing and it is important as it has been used in various fields such as medical imaging, autonomous driving and object detection. The selection of technique that is suitable for application is crucial in order to get a good result. The problem occurs when one would like to choose the right method that will give a good segmented image. Therefore, this paper presents the image quality assessment for image segmentation algorithms in terms of qualitative and quantitative analyses. In this project, three techniques of image segmentation which are K-means clustering, threshold and watershed marker controlled are compared with each other and the evaluation result is based on the image quality assessment. Those three methods used to segment greyscale images are taken from the Internet. The objective is to find which technique gives the best result based on the image quality assessment. The parameters used to analyse the quality of the processed image are the mean square error (MSE), average difference (AD), mean absolute error (MAE), structural similarity index metric (SSIM), also the structural dissimilarity index metric (DSSIM). The algorithm of the image segmentation techniques is built using open CV-C++. Based on the assessment, the results, it shows that K-means clustering gives a better result qualitatively and quantitatively while the watershed marker-based shows a good qualitative result. This quality assessment technique hopefully could help in the selection of the image segmentation technique for the respective applications.