Research Progress of Image Segmentation

Abstract

Image segmentation is an essential and challenging task in CV. It can be used in a massive scope including automatic driving, medical analysis and geological detection. In the early stage of the research, some researchers based on the geometry of the image, color space and other ways of semantic segmentation results are poor. Fortunately, with the flourishing of deep learning, great improvement has been gained in various deep learning-based image segmentation fields. At the same time, in the open world, semantic segmentation research is not limited to fixed categories or known categories, and some open-world semantic segmentation methods and reference semantic segmentation methods based on text description have also made great breakthroughs. This paper summarizes commonly used datasets for segmentation and introduces main ideas of popular algorithms in traditional, open-vocabulary and referring segmentation respectively. In the end, the important contribution of the paper is summarized and some suggestions regarding the future improvement of existing algorithms are given.