Cluster Sensing Superpixel and Grouping

Rui Li, Lu Fang; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2016, pp. 46-54

Abstract


Superpixel algorithms have shown significant potential in computer vision applications since they can be used to accelerate other computationally demanding algorithms.However, in contrast to the original purpose of superpixels, many upper layer methods still suffer from computational problems when incorporating superpixel for speedup. In this paper, we present a cluster sensing superpixel (CSS) method to efficiently generate superpixel bricks. Based on the insight of pixel density, cluster centers generally have properties of representativeness (i.e., local maximal pixel density) and isolation (i.e., large distance from other cluster centers). Our CSS method efficiently identifies ideal cluster centers via utilizing pixel density. We also integrate superpixel cues into a bipartite graph segmentation framework and apply it to microscopy image segmentation. Extensive experiments show that our CSS method achieves impressive efficiency, being approximately five times faster than the state-of-the-art methods and having comparable performance in terms of the standard metrics. Application on microscopy image segmentation also benefits our efficient implementation.

Related Material


[pdf]
[bibtex]
@InProceedings{Li_2016_CVPR_Workshops,
author = {Li, Rui and Fang, Lu},
title = {Cluster Sensing Superpixel and Grouping},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2016}
}