Spectral Segmentation with Multiscale Graph Decomposition

Abstract

We present a multiscale spectral image segmentation algorithm. In contrast to most multiscale image processing, this algorithm works on multiple scales of the image in parallel, without iteration, to capture both coarse and fine level details. The algorithm is computationally efficient, allowing to segment large images. We use the Normalized Cut graph partitioning framework of image segmentation. We construct a graph encoding pairwise pixel affinity, and partition the graph for image segmentation.We demonstrate that large image graphs can be compressed into multiple scales capturing image structure at increasingly large neighborhood. We show that the decomposition of the image segmentation graph into different scales can be determined by ecological statistics on the image grouping cues. Our segmentation algorithm works simultaneously across the graph scales, with an inter-scale constraint to ensure communication and consistency between the segmentations at each scale. As the results show, we incorporate long-range connections with linear-time complexity, providing high-quality segmentations efficiently. Images that previously could not be processed because of their size have been accurately segmented thanks to this method. This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of the University of Pennsylvania's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE by writing to pubs-permissions@ieee.org. By choosing to view this document, you agree to all provisions of the copyright laws protecting it.This conference paper is available at ScholarlyCommons: http://repository.upenn.edu/cis_papers/218 Spectral Segmentation with Multiscale Graph Decomposition

TurboPixels: Fast Superpixels Using Geometric Flows

“…The TurboPixels algorithm was implemented in Matlab with several C extensions. 4 For N-cuts, we use the 2004 N-cut implementation based on [27], 5 while for Sb, we simply divide the image into even rectangular blocks, providing a naive but efficient benchmark for accuracy (other algorithms are expected to do better). All experiments were performed on a quadcore Xeon 3.6 GHz computer.…”

“…However, the number of superpixels is no longer directly controlled nor is the algorithm designed to ensure the quasi uniformity of segment size and shape. Cour et al [4] also proposed a linear time algorithm by solving a constrained multiscale N-Cuts problem, but this complexity does not take the number of superpixels into account. In practice, this method remains computationally expensive and thus unsuitable for large images with many superpixels.…”

Simultaneous Segmentation and Figure/Ground Organization Using Angular Embedding

“…Figure 9 compares our automatically generated segmentations to those of other algorithms [8,7,6,2,3] using the standard BSDS boundary precision-recall benchmark [18]. Precision-recall curves for the other algorithms are those reported in [3].…”

“…Evaluation of region boundaries on the BSDS Benchmark. Left: The segmentation quality of our algorithm is close to that of the current best-performing algorithm, gP b-owt-ucm [3], and superior to others [8,7,6,2], as benchmarked by [3]. Algorithms are evaluated in terms of precision and recall with respect to human groundtruth boundaries.…”

A human visual system-driven image segmentation algorithm

“…These include Gaussian assumption-based dynamic clustering algorithm [29] (GADCA), iterative mode separation algorithm [29] (IMSA), higher-order statistics method based on local maxima detection and adaptive wavelet transform [30] (HOSLW), the conventional MRF-based algorithm, Otsu thresholding [31] [32], the level set evolution-based method without reinitialization (LSEWRI) [33], the region-based active contour model (RACM) [34], and the multi-scale normalized cuts-based segmentation (MNCut) [35]. Experiments and analyses are carried out on medical and natural images, where the medical images are two typical mammogram and MRI brain images, and the Cameraman image is used as the representative image of a natural scene.…”