Image Segmentation Using Hierarchical Merge Tree

Abstract

This paper investigates one of the most fundamental computer vision problems: image segmentation. We propose a supervised hierarchical approach to object-independent image segmentation. Starting with over-segmenting superpixels, we use a tree structure to represent the hierarchy of region merging, by which we reduce the problem of segmenting image regions to finding a set of label assignment to tree nodes. We formulate the tree structure as a constrained conditional model to associate region merging with likelihoods predicted using an ensemble boundary classifier. Final segmentations can then be inferred by finding globally optimal solutions to the model efficiently. We also present an iterative training and testing algorithm that generates various tree structures and combines them to emphasize accurate boundaries by segmentation accumulation. Experiment results and comparisons with other recent methods on six public data sets demonstrate that our approach achieves state-of-the-art region accuracy and is competitive in image segmentation without semantic priors.

Cell Detection Using Extremal Regions in a Semisupervised Learning Framework

“…if y i = 1, y i ′ = 0, and i ′ ∈ D i , where D i is the set of descendent nodes i . The optimization problem is solved in a similar manner as seen in [20]. Using the tree structure, we use dynamic programming to find the best and most efficient solution with the path consistency loss.…”

Topologically-Guided Color Image Enhancement

“…For example, algorithms for edge detection and image segmentation detected "contours" (i.e., edges) in a hierarchical manner for segmentation [1]. Supervised learning, such as the hierarchical merge trees model in [12], are popular in image segmentation. Contrast enhancement has been proposed as an optimization problem that maximizes the average local contrast of an image [13].…”

Cell segmentation in 3D confocal images using supervoxel merge-forests with CNN-based hypothesis selection

“…Supervoxel-based segmentation approaches have been successfully applied in various computer vision domains (including segmentation of 2D and 3D microscopy images) and form the basis of the method presented in this contribution [6,8,13,16]. The approaches consist of mainly three steps: (1) a preprocessing stage to improve the image signal and to enhance the desired object boundaries, (2) a supervoxel generation step that partitions the enhanced image into meaningful parts and (3) a supervoxel merging phase that tries to agglomerate existing supervoxels to the desired objects of interest.…”

“…Each node of a merge-tree corresponds to a potential segmentation hypothesis. In contrast to existing methods that precompute a single merge-tree for the entire image [3,6,8], we terminate the merging phase based on a specimen-dependent maximum volume constraint. This allows us to pick the best hypotheses using a CNN-based postprocessing step that only needs to be trained on small image snippets that maximally cover a few cells rather than all levels of detail of the entire image.…”