Hierarchical Region Merging for Multi-scale Image Segmentation

Abstract

Image segmentation is a key computer vision technique that divides the pixels of an image into different blocks of distinct transactions. The multi-scale segmentation method is one of the image segmentation methods, which can extract the object regions of different scales. It has the potential to fully exploit the application of high resolution and complex scene images and is the research hotspots direction of image segmentation technology. In this work, a feasible image scale-aware algorithm is proposed. By using the segmentation results of the existing multi-scale segmentation algorithm, the global region’s hierarchical region is merged by the quantitative description of each hierarchical region feature to achieve the optimal scale of multi-scale segmentation. We validate the proposed method on different algorithms and data sets. The results have shown that the proposed method can solve the error caused by manual threshold setting and achieve the optimal selection of individual goals to a certain extent.