Interactive Image Segmentation Based on Synthetic Graph Coordinates

Introduction

This work is focused to propose a framework for interactive image segmentation problem. The goal of interactive image segmentation is to classify the image pixels into foreground and background classes, when some foreground and the background markers are given (see Figure 1).

Figure 1: (a) Original image, (b), (c) given markers and (d) the ground truth image.

Experimental results and comparisons with other methods from literature are presented on several data sets like LHI, Gulshan and Zhao demonstrating the high-performance of the proposed scheme [5].

Methodology

Figure 2: Scheme of the proposed system architecture.

The proposed scheme [5] consists of several steps.

    • First, we partition the image into superpixels, that are contiguous and perceptually similar regions.
    • Then, we construct a weighted graph that represents the superpixels and the connections between them, taking into account the given markers and visual information.
    • An efficient algorithm for graph clustering based on synthetic coordinates is used yielding an initial map of classified pixels.
    • Finally, having available the data modeling and the initial map of classified pixels, we used a Markov Random Field (MRF)
    • model or a flooding algorithm getting the image segmentation.

Synthetic coordinates [1-6]

    • Each node gets a position in high-dimensional Euclidean space using Vivaldi algorithm [7]. Vivaldi is a fully decentralized, light-weight, adaptive network coordinate algorithm that predicts Internet latencies with low error.
    • Vivaldi simulates a network of physical springs. We place imaginary springs between selected pairs of connected nodes (traction) and non-connected nodes (turning away).

Figure 3. The two types of physical springs for connected nodes (traction) and non-connected nodes (turning away) .

Downloads

    • You can download the experimental results of the proposed method [5] on the following three data setsLHI, Gulshan and Zhao. See the corresponding readme.txt files for instructions. You can use them only for non-commercial purposes. If you use them, please cite the article [5].

Related Publications

[1] I. Grinias, N. Komodakis and G. Tziritas, Flooding and MRF-based Algorithms for Interactive Segmentation,

Intern. Conf. on Pattern Recognition, 2010.

[2]. H. Papadakis, C. Panagiotakis and P. Fragopoulou, Locating Communities on Real Dataset Graphs Using Synthetic Coordinates, Parallel Processing Letters (PPL), vol. 22, no. 1, Jan. 2012.

[3]. H. Papadakis, C. Panagiotakis and P. Fragopoulou, Locating Communities on Graphs with variations in Community Sizes, Journal of Supercomputing, vol. 185, no.1, pp. 9-15, Jan. 2012.

[4]. H. Papadakis, C. Panagiotakis and P. Fragopoulou, Distributed Community Detection in a Complex World Using Synthetic Coordinates, submitted to PAMI, 2012.

[5]. H. Papadakis, C. Panagiotakis and P. Fragopoulou, Interactive Image Segmentation Based on Synthetic Graph Coordinates, submitted to Pattern Recognition, 2012.

[6]. H. Papadakis, C. Panagiotakis and P. Fragopoulou, Local Community Finding using Synthetic Coordinates, International Conference on Future Information Technology (FutureTech 2011), 2011 (Best Paper Award).

[7]. Frank Dabek, Russ Cox, Frans Kaashoek, and Robert Morris. Vivaldi: A decentralized network coordinate system. In Proceedings of the ACM SIGCOMM ’04 Conference, August 2004.

Acknowledgments: This work is partially supported by the ``Thalis'' (Project's Acronym: UrbanMonitor) and ``ARCHIMEDE III: Education and Lifelong Learning'' (Project's Acronym: P2PCOORD) projects.