Hierarchical Image Segmentation Based on Bi-Directional Cascade Network

Image segmentation is the way of splitting digital data into subgroups known as image segments, which reduces the entire image's complexity and allows for further processing or image analysis of each segment. The general segmentation techniques of images are divided into traditional methods and deep learning methods. In this paper, we introduce a deep-learning approach to a traditional image segmentation method (i.e., edge detection) and propose a new framework for hierarchical image segmentation. In the new framework, we first use the Bi-Directional Cascade Network (BDCN) model as our edge detector to preprocess the input image in order to form a feature map with coarse edges. Then, we refine the edges of the image by using the watershed algorithm. Finally, we form a segmentation dendrogram by using the Ultrametric Contour Map (UCM) algorithm. Extensive experiments have demonstrated that our proposed framework has significant advantages in segmentation performance compared to the state-of-the-art algorithms.