EM/MPM

Group (Subgroup)

Reconstruction (Segmentation)

Description

This Filter performs the EM/MPM segmentation algorithm on an Attribute Array representing a grayscale image. The EM/MPM algorithm employs an advanced expectation maximization routine over Gaussian mixtures to determine an image segmeneation into a defined number of classes. The segmented image will be stored into a new Attribute Array with a user definable name. Note that the created segmentation will have Cell labels defining the class membership. Thus, the labels will be unsigned 8 bit integers, matching the incoming grayscale image. These labels can be considered Feature Ids for the purposes of most DREAM.3D analysis routines. However, DREAM.3D assumes that Feature Ids are signed 32 bit integers. It may therefore be required to use the Convert Attribute Data Type Filter to convert the segmented image labels from unsigned 8 bit integers to signed 32 bit integers for further analysis.

It is highly recommended that users consult references [1], [2], [3], and [4] for details on the impact of particular parameters on the EM/MPM algorithm.

Parameters

Name Type Description
Number of Classes int32_t The number of classes in which to segment the image
Exchange Energy float The value of the exchange energy
Histogram Loops (EM) int32_t The number of histogram loops (EM) to perform
Segmentation Loops (MPM) int32_t The number of segmentation loops (MPM) to perform
Use Simulated Annealing bool Apply the simulated annealing process
Use Gradient Penalty bool Use a penalty to gradients when segmenting
Gradient Penalty (Beta E) float The penalty to apply for gradients. Only needed if Use Gradient Penalty is checked
Use Curvature Penalty bool Use a penalty for curvature
Curvature Penalty float The penalty to use for curvatures. Only needed if Use Curvature Penalty is checked
R Max float The max radius for the curvature penalty. Only needed if Use Curvature Penalty is checked
EM Loop Delay int32_t The number of EM Loops to delay before applying the curvature penalty. Only needed if Use Curvature Penalty is checked
Use 1-Based Values bool Use 1-based values instead of 0-based values

Required Geometry

Image

Required Objects

Kind Default Name Type Component Dimensions Description
Cell Attribute Array None uint8_t (1) 8 bit grayscale image to segment

Created Objects

Kind Default Name Type Component Dimensions Description
Cell Attribute Array None uint8_t (1) Unsigned 8 bit array representing the value of the class that the Cell was segmented into

References

[1] H.C. Chuang, M.L. Comer, J.P. Simmons, Texture Segmentation in Microstructure Images of Advanced Materials, In Proceedings of the 2008 IEEE Southwest Symposium on Image Analysis and Interpretation, March 24-26, 2008, Santa Fe, NM

[2] M.L. Comer, The EM/MPM Algorithm for Segmentation of Textured Images: Analysis and Further Experimental Results, IEEE Transactions on Image Processing, vol. 9, no. 10, October 2000.

[3] J.P. Simmons, P. Chuang, M. Comer, J.E. Spowart, M.D. Uchic, M. De Graef, Application and further development of advanced image processing algorithms for automated analysis of serial section image data, Modelling Simul. Mater. Sci. Eng. 17 (2009).

[4] D. Doria, Expectation Maximization of Gaussian Mixture Models in VTK - Release 0.00, September 21, 2010, Insight Journal http://hdl.handle.net/10380/3218

Example Pipelines

Please see the description file distributed with this Plugin

DREAM.3D Mailing Lists

If you need more help with a Filter, please consider asking your question on the DREAM.3D Users Google group!