Image segmentation using fuzzy min-max neural networks for wood defect detection

G.A. Ruza,b, and P.A. Estéveza

a Department of Electrical Engineering, University of Chile, Chile
b SONDA S.A., Chile

Outline of the presentation

fuzzy min-max neural networks

f

Clustering example using FMM NN 

f2

AVI system for wood boards

Avi

segmentation module 

Seg_Mod

image segmentation method

imag_seg_method1

Seed selection process

  • To speed up the image segmentation process, the FMMIS does not use all the pixels from the image analysed. Instead, it only uses a few input pixels called seeds, to grow the hyperboxes.

  • The seeds are automatically determined (located) by an ad-hoc procedure, since this process is problem-dependent.

  • Considering the great variability of colour of the wood boards, the seed selection is based on adaptive thresholding in the RGB colour space.

input patterns

  • The input patterns are the spatial coordinates of the seeds, with each dimension normalized in the range [0,1].

  • Let X be a S x 2 input matrix, where S is the number of seeds selected

  • The position of the hth seed in the image is represented by the vector Xh = (xh1,xh2), where the first coordinate indicates the column and the second coordinate the row of the image.

Fuzzy min-max neural network for image segmentation (FMMIS)

  • The FMMIS method places hyperboxes defined in the 2D geometric space by pairs of min-max points for each spatial coordinate of the image (rectangular boxes in the case of 2D images).
  • Each hyperbox fuzzy set has an associated membership function that describes the degree of membership (spatial proximity) of a given pixel to a hyperbox in the [0,1] interval.
  • When an input pattern (new seed) is presented, the hyperbox with the highest degree of membership is found and expanded to enclose the input pattern.
  • The hyperbox expansion is accepted only if the region contained by the expanded hyperbox is similar in colour to the region enclosed by the hyperbox before the expansion. 

FMMIS learning algorithm

  1. Initialisation
  2. Hyperbox Expansion
  3. Hyperbox Ovelap Test
  4. Hyperbox Contraction
  5. Fine-tuning Hyperbox Expansion
  6. Hyperbox Merging

output of the fMMIS

output2

methods

db

experimental results

results

Some examples

examples 

conclusions

  • The proposed colour image segmentation method achieved a high defect detection rate (95%) with a low false positive rate (6%) on images of wood boards.
  • The FMMIS method is based on the original FMM, but with a new learning algorithm specially adapted for image segmentation tasks.
  • The FMMIS method combines clustering with region-based techniques to obtain a substantially different method than the original Simpson’s FMM.
  • The results show that significant improvements have been obtained in comparison to previous work.