Research and Practice of Unsupervised Statistical Methods in Image Segmentation

Image segmentation is a basic task of computer vision, it is to divide the picture into various fragments and objects. Supervised deep learning, it has reached SOTA but its dependance on large Scale Annotated Data Set is limit. unsupervised methods or methods based on statistics give an interesting alternative in finding an interesting pattern within the data itself as opposed to manually placing the data into human provided labels. Paper gives an all-round review and analysis on the unsupervised statistical method for image segmentation. So we will now split into systematically ways clustering based ways regions based ways as well as Probabilistic models. Regarding every category, we go deep into its main statistical assumptions, algorithmic procedures, and actual usage. Apart from these, there is a sort of experimentation on various kinds of dataset with a comparison analysis and measuring the result as per certain standards. the results attained highlight the special strengths and limits of each approach, offering knowledge concerning how well suited they are for various imaging scenarios. Finally, we mention ongoing issues and propose promising new avenues for statistical language models, which include building hybrid models that combine the statistics and deep feature representations.