Boundary Extraction in Natural Images Using Ultrametric Contour Maps

By Pablo Arbeláez. 2006

253 citations recorded by the source corpus.

This paper presents a low-level system for boundary extraction and segmentation of natural images and the evaluation of its performance. We study the problem in the framework of hierarchical classification, where the geometric structure of an image can be represented by an ultrametric contour map, the soft boundary image associated to a family of nested segmentations. We define generic ultrametric distances by integrating local contour cues along the regions boundaries and combining this information with region attributes. Then, we evaluate quantitatively our results with respect to ground-truth segmentation data, proving that our system outperforms significantly two widely used hierarchical segmentation techniques, as well as the state of the art in local edge detection.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Open-World Instance Segmentation: Exploiting Pseudo Ground Truth From Learned Pairwise Affinity (2022, methodology): “We found that UCM-based methods significantly outperforms other two methods (Table 2): whereas CC and GBH make decision on a merge based on a single pairwise relationship, UCM uses all relations between two segments, and is more robust.”
Multiscale Combinatorial Grouping for Image Segmentation and Object Proposal Generation (2015, background): “thermore, it can also be represented as an ultrametric contour map (UCM), an image obtained by weighting the boundary of each pair of adjacent regions in the hierarchy by the index at which they are merged [31], [32].”
Segmentation of Natural Images by Texture and Boundary Compression (2010, unknown): “To quantitatively evaluate the performance of our method, we use three metrics for comparing pairs of image segmentations: the probabilistic Rand index (PRI) (Rand (1971)), the variation of information (VOI) (Meila (2005)), and the global F-measure (Arbelaez (2006)): 1.”
Multi-scale Improves Boundary Detection in Natural Images (2008, background): “68, higher than most existing results on the Berkeley benchmark [22, 23, 11, 24, 25, 12].”
Maximin affinity learning of image segmentation (2009, methodology): “The Ultrametric Contour Map algorithm [1] generates hierarchical segmentations nearly identical those generated by varying the threshold of our graph partitioning algorithm.”
Recovering Occlusion Boundaries from a Single Image (2007, methodology): “However, most segmentation algorithms rely on 2D perceptual grouping cues, such as brightness, color, or texture similarity for region-based methods [1, 6, 18] or edge strength, continuity, and closure for contour-based methods (e.g., [14]).”
Bayesian Semantic Instance Segmentation in Open Set World (2018, methodology): “A region hierarchy can be eﬃciently represented using a single Ultrametric Contour Map (UCM) [1].”
Multiscale Combinatorial Grouping (2014, background): “Furthermore, it can also be represented as an ultrametric contour map (UCM), an image obtained by weighting the boundary of each pair of adjacent regions in the hierarchy by the index at which they are merged [2].”
Boundary Detection Benchmarking: Beyond F-Measures (2013, background): “Their alarming results suggest the potential pitfalls of some widely adopted benchmarks.”
Boundary Detection Using F-Measure-, Filter- and Feature- (F3) Boost (2010, background): “This approach has led to improved detectors based on Boosting [6], topological properties of the image [7], spectral gradients [8], multiscale processing [9] and sparse dictionaries [10], among others.”
Learning full pairwise affinities for spectral segmentation (2010, methodology): “Hierarchical segmentation: Unlike the state-of-the-art learning-based methods mPb-OWT-UCM [35] and gPb-OWT-UCM [33], our algorithm fPb-OWT-UCM works without training.”
Instance Segmentation of Indoor Scenes Using a Coverage Loss (2014, methodology): “While various schemes [21] [10] have been introduced to perform this operation, we use the method of [22].”
Using contours to detect and localize junctions in natural images (2008, background): “A thinned, real-valued contour image can be obtained from gPb by considering the maximal response over orientations at each location and applying non-maximum suppression [3, 12].”
Affinity CNN: Learning Pixel-Centric Pairwise Relations for Figure/Ground Embedding (2015, background): “For example, image segmentation follows from edge detection [12, 1, 2, 7, 4]; figure/ground, occlusion, or depth layering follows from reasoning over discrete contours or regions [27, 16, 21, 36, 18] with some systems also reliant on motion cues [30, 15, 32, 31].”
From contours to regions: An empirical evaluation (2009, methodology): “We then construct an Ultrametric Contour Map (UCM) [2] from the boundaries of these initial regions.”
Discriminative Sparse Image Models for Class-Specific Edge Detection and Image Interpretation (2008, methodology): “Precision-recall curves are presented on Figure 4 and are compared with Pb [18], BEL [5], UCM [2] and the recent gPb [16], which was published after that this paper was accepted.”
Self-Supervised Visual Representation Learning from Hierarchical Grouping (2020, unknown): “This is precisely the procedure for constructing an ultrametric contour map (UCM) [1], which can equivalently be regarded as as both a reweighted edge map, and a tree T deﬁning a region hierarchy.”
Contour Detection and Hierarchical Image Segmentation (2011, methodology): “17 are pairs of curves corresponding to contour detector output and regions produced by running the OWT-UCM algorithm on that output.”