Random walker algorithm

From Wikipedia, the free encyclopedia
Jump to: navigation, search

The random walker algorithm is an algorithm for image segmentation. In the first description of the algorithm,[1] a user interactively labels a small number of pixels with known labels (called seeds), e.g., "object" and "background". The unlabeled pixels are each imagined to release a random walker, and the probability is computed that each pixel's random walker first arrives at a seed bearing each label, i.e., if a user places K seeds, each with a different label, then it is necessary to compute, for each pixel, the probability that a random walker leaving the pixel will first arrive at each seed. This computation may be determined analytically by solving a system of linear equations. After computing these probabilities for each pixel, the pixel is assigned to the label for which it is most likely to send a random walker. The image is modeled as a graph, in which each pixel corresponds to a node which is connected to neighboring pixels by edges, and the edges are weighted to reflect the similarity between the pixels. Therefore, the random walk occurs on the weighted graph (see Doyle and Snell for an introduction to random walks on graphs[2]).

Although the initial algorithm was formulated as an interactive method for image segmentation, it has been extended to be a fully automatic algorithm, given a data fidelity term (e.g., an intensity prior).[3] It has also been extended to other applications, such as Image Matching (R. Shen, I. Cheng, X.li and A. Basu), ICPR 2008, and Image Fusion, (R. Shen, I. Cheng, J.Shi and A. Basu), IEEE Trans. on Image Processing, 2011, and other applications.

The algorithm was initially published as a conference paper[4] and later as a journal paper.[1]

Mathematics[edit]

Although the algorithm was described in terms of random walks, the probability that each node sends a random walker to the seeds may be calculated analytically by solving a sparse, positive-definite system of linear equations with the graph Laplacian matrix, which we may represent with the variable L. The algorithm was shown to apply to an arbitrary number of labels (objects), but the exposition here is in terms of two labels (for simplicity of exposition).

Assume that the image is represented by a graph, with each node v_i associated with a pixel and each edge e_{ij} connecting neighboring pixels v_i and v_j. The edge weights are used to encode node similarity, which may be derived from differences in image intensity, color, texture or any other meaningful features. For example, using image intensity g_i at node v_i, it is common to use the edge weighting function

w_{ij} = \exp{\left(-\beta (g_i - g_j)^2\right)}.

The nodes, edges and weights can then be used to construct the graph Laplacian matrix.

The random walker algorithm optimizes the energy

Q(x) = x^T L x = \sum_{e_{ij}} w_{ij} \left(x_i - x_j\right)^2

where x_i represents a real-valued variable associated with each node in the graph and the optimization is constrained by x_i = 1 for v_i \in F and x_i = 0 for v_i \in B, where F and B represent the sets of foreground and background seeds, respectively. If we let S represent the set of nodes which are seeded (i.e., S = F \cup B) and \overline{S} represent the set of unseeded nodes (i.e., S \cup \overline{S} = V where V is the set of all nodes), then the optimum of the energy minimization problem is given by the solution to


L_{\overline{S},\overline{S}} x_{\overline{S}} = - L_{\overline{S},S} x_{S},

where the subscripts are used to indicate the portion of the graph Laplacian matrix L indexed by the respective sets.

To incorporate likelihood (unary) terms into the algorithm, it was shown in [3] that one may optimize the energy

Q(x) = x^T L x  + \gamma \left((1-x)^T F (1-x) + x^T B x\right) = \sum_{e_{ij}} w_{ij} \left(x_i - x_j\right)^2 + \gamma \left(\sum_{v_i} f_i (1-x_i)^2 + \sum_{v_i} b_i x_i^2 \right),

for positive, diagonal matrices F and B. Optimizing this energy leads to the system of linear equations


\left(L_{\overline{S},\overline{S}} + \gamma F_{\overline{S},\overline{S}} + \gamma B_{\overline{S},\overline{S}}\right) x_{\overline{S}} = - L_{\overline{S},S} x_{S} - \gamma F_{\overline{S},\overline{S}}.

The set of seeded nodes, S, may be empty in this case (i.e., \overline{S}=V), but the presence of the positive diagonal matrices allows for a unique solution to this linear system.

For example, if the likelihood/unary terms are used to incorporate a color model of the object, then f_i would represent the confidence that the color at node v_i would belong to object (i.e., a larger value of f_i indicates greater confidence that v_i belonged to the object label) and b_i would represent the confidence that the color at node v_i belongs to the background.

Algorithm interpretations[edit]

The random walker algorithm was initially motivated by labeling a pixel as object/background based on the probability that a random walker dropped at the pixel would first reach an object (foreground) seed or a background seed. However, there are several other interpretations of this same algorithm which have appeared in.[1]

Circuit theory interpretations[edit]

There are well-known connections between electrical circuit theory and random walks on graphs.[5] Consequently, the random walker algorithm has two different interpretations in terms of an electric circuit. In both cases, the graph is viewed as an electric circuit in which each edge is replaced by a passive linear resistor. The resistance, r_{ij}, associated with edge e_{ij} is set equal to r_{ij} = \frac{1}{w_{ij}} (i.e., the edge weight equals electrical conductance).

In the first interpretation, each node associated with a background seed, v_i \in B, is tied directly to ground while each node associated with an object/foreground seed, v_i \in F is attached to a unit direct current ideal voltage source tied to ground (i.e., to establish a unit potential at each v_i \in F). The steady-state electrical circuit potentials established at each node by this circuit configuration will exactly equal the random walker probabilities. Specifically, the electrical potential, x_i at node v_i will equal the probability that a random walker dropped at node v_i will reach an object/foreground node before reaching a background node.

In the second interpretation, labeling a node as object or background by thresholding the random walker probability at 0.5 is equivalent to labeling a node as object or background based on the relative effective conductance between the node and the object or background seeds. Specifically, if a node has a higher effective conductance (lower effective resistance) to the object seeds than to the background seeds, then node is labeled as object. If a node has a higher effective conductance (lower effective resistance) to the background seeds than to the object seeds, then node is labeled as background.

Extensions[edit]

The traditional random walker algorithm described above has been extended in several ways:

Applications[edit]

Beyond image segmentation, the random walker algorithm has been additionally applied to several problems in computer vision and graphics:

References[edit]

  1. ^ a b c L. Grady: Random Walks for Image Segmentation, IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. 28, No. 11, pp. 1768–1783, Nov., 2006.
  2. ^ P. Doyle, J. L. Snell: Random Walks and Electric Networks, Mathematical Association of America, 1984
  3. ^ a b Leo Grady: Multilabel Random Walker Image Segmentation Using Prior Models, Proc. of CVPR, Vol. 1, pp. 763–770, 2005. [1]
  4. ^ Leo Grady, Gareth Funka-Lea: Multi-Label Image Segmentation for Medical Applications Based on Graph-Theoretic Electrical Potentials, Proc. of the 8th ECCV Workshop on Computer Vision Approaches to Medical Image Analysis and Mathematical Methods in Biomedical Image Analysis, pp. 230–245, 2004.
  5. ^ P. G. Doyle, J. L. Snell: Random Walks and Electrical Networks, Carus Mathematical Monographs, 1984
  6. ^ T. H. Kim, K. M. Lee, S. U. Lee: Generative Image Segmentation Using Random Walks with Restart, Proc. of ECCV 2008, pp. 264–275
  7. ^ J. Wang, M. Agrawala, M. F. Cohen: Soft scissors: an interactive tool for realtime high quality matting, Proc. of SIGGRAPH 2007
  8. ^ S. Rysavy, A. Flores, R. Enciso, K. Okada: Classifiability Criteria for Refining of Random Walks Segmentation, Proc. of ICPR 2008
  9. ^ W. Yang, J. Cai, J. Zheng, J. Luo: User-friendly Interactive Image Segmentation through Unified Combinatorial User Inputs, IEEE Trans. on Image Proc., 2010
  10. ^ C. Chefd'hotel, A. Sebbane: Random walk and front propagation on watershed adjacency graphs for multilabel image segmentation, Proc. of ICV 2007
  11. ^ R. Rzeszutek, T. El-Maraghi, D. Androutsos: Image segmentation using scale-space random walks, Proc. of the 16th international conference on Digital Signal Processing, pp. 458–461, 2009
  12. ^ L. Grady, A.K. Sinop: Fast approximate random walker segmentation using eigenvector precomputation. In IEEE Conf. CVPR, pp. 1–8, 2008
  13. ^ S. Andrews, G. Hamarneh, A. Saad. Fast random walker with priors using precomputation for interactive medical image segmentation, Proc. of MICCAI 2010
  14. ^ X. Liu, J. Liu, Z. Feng: Colorization Using Segmentation with Random Walk, Computer Analysis of Images and Patterns, pp. 468–475, 2009
  15. ^ R. Rzeszutek, T. El-Maraghi, D. Androutsos: Interactive rotoscoping through scale-space random walks, Proc. of the 2009 IEEE international conference on Multimedia and Expo
  16. ^ S. P. Dakua, J. S. Sahambi: LV Contour Extraction from Cardiac MR Images Using Random Walks Approach, Int. Journal of Recent Trends in Engineering, Vol 1, No. 3, May 2009
  17. ^ F. Maier, A. Wimmer, G. Soza, J. N. Kaftan, D. Fritz, R. Dillmann: Automatic Liver Segmentation Using the Random Walker Algorithm, Bildverarbeitung für die Medizin 2008
  18. ^ P. Wighton, M. Sadeghi, T. K. Lee, M. S. Atkins: A Fully Automatic Random Walker Segmentation for Skin Lesions in a Supervised Setting, Proc. of MICCAI 2009
  19. ^ P. Wattuya, K. Rothaus, J. S. Prassni, X. Jiang: A random walker based approach to combining multiple segmentations, Proc. of ICPR 2008
  20. ^ Y.-K. Lai, S.-M. Hu, R. R. Martin, P. L. Rosin: Fast mesh segmentation using random walks, Proc. of the 2008 ACM symposium on Solid and physical modeling
  21. ^ J. Zhang, J. Zheng, J. Cai: Interactive Mesh Cutting Using Constrained Random Walks, IEEE Trans. on Visualization and Computer Graphics, 2010.
  22. ^ X. Sun, P. L. Rosin, R. R. Martin, F. C. Langbein: Random walks for feature-preserving mesh denoising, Computer Aided Geometric Design, Vol. 25, No. 7, Oct. 2008, pp. 437–456
  23. ^ L. Grady, G. Funka-Lea: An Energy Minimization Approach to the Data Driven Editing of Presegmented Images/Volumes, Proc. of MICCAI, Vol. 2, 2006, pp. 888–895
  24. ^ G. Li, L. Qingsheng, Q. Xiaoxu: Moving Vehicle Shadow Elimination Based on Random Walk and Edge Features, Proc. of IITA 2008
  25. ^ R. Shen, I. Cheng, X. Li, A. Basu: Stereo matching using random walks, Proc. of ICPR 2008
  26. ^ R. Shen, I. Cheng, J. Shi, A. Basu: Generalized Random Walks for Fusion of Multi-exposure Images, IEEE Trans. on Image Processing, 2011.

External links[edit]