Training Image Operators from Samples - Research at IME-USP

Image Processing can be used to solve complex problems in many different areas, such as document analysis and face detection. However, the design of image processing operators is a very challenging task that requires deep knowledge of both image processing techniques and the domain of application. An alternative formulation consists in using machine learning to estimate a local transformation from a set of pairs of images containing an input and its processed version. More details about this formulation can be found at our Introduction to Image Operator Learning.

This page concentrates the research done on Image Operator Learning at the eScience group of the Institute of Mathematics and Statistics of the University of São Paulo.




Other contributors:


One of our most recent contributions is the development of TRIOSlib, an image operator learning library that exposes many methods of the area so that other researchers can also benefit from our work.

Find more details at TRIOSlib's page on SourceForge or download the code at the project's page.


This page is also home to the supplementary material of papers related to TRIOSlib.

  1. I. S. Montagner, R. Hirata Jr, N. S. T. Hirata, "A Machine Learning based method for Staff Removal", accepted for publication at ICPR 2014. (supplementary material)
  2. I. S. Montagner, R. Hirata Jr, N. S. T. Hirata, "Learning to Remove Staff Lines from Music Score Images", accepted for publication at ICIP 2014. (supplementary material)
  3. N. S. T. Hirata, "Multilevel training of binary morphological operators", IEEE Transactions on Pattern Analysis and Machine Intelligence, Volume 31, Issue 4, pp.707-720, April 2009. (supplementary material)

Other related papers by our group are :

  1. M.M. Dornelles and N. S. T. Hirata, A genetic algorithm based approach for combining binary image operators. 21st International Conference on Pattern Recognition, 2012, Tsukuba.
  2. HIRATA, N. S. T. . Morphological Operator Design from Training Data: A State of the Art Overview. In: Halina Kwasnicka; Lakhmi C. Jain. (Org.). Innovations in Intelligent Image Analysis, Springer, 2011, p. 31-58.
  3. Carlos S. Santos, Nina S.T. Hirata, and Roberto Hirata Jr. "An information theory framework for two-stage binary image operator design". Pattern Recognition Letters, 31(4):297-306, 2010.
  4. N. S. T. Hirata, R. Hirata Jr., and J. Barrera. "Basis computation algorithms". In Mathematical Morphhology and its Applications to Signal and Image Processing (Proceedings of the 8th International Symposium on Mathematical Morphology), pages 15-26, 2007.
  5. D. Dellamonica Jr., P. J. S. Silva, C. Humes Jr., N. S. T. Hirata, and J. Barrera, “An Exact Algorithm for Optimal MAE Stack Filter Design,” IEEE Transactions on Image Processing, vol. 16, no. 2, pp. 453–462, 2007.
  6. HIRATA JR., R. ; BRUN, M. ; BARRERA, J. ; DOUGHERTY, E. R.. Aperture Filters: Theory, Application, and Multiresolution Analysis. Em: Stephen Marshall;Giovanni L. Sicuranza. (Org.). Advances on Nonlinear Signal and Image Processing. : EURASIP Book Series on Signal Processing and Communications. 2007.
  7. Marcel Brun, Roberto Hirata Jr., Junior Barrera, and Edward R. Dougherty. Nonlinear Filter design using envelopes. Journal of Mathematical Imaging and Vision, 21(1-2):81-97, 2004.
  8. J. Barrera, R. Terada, R. Hirata Jr, N. S. T. Hirata, Automatic programming of morphological machines by PAC learning. Fundamenta Informaticae 41 (1), 229-258


Download below some datasets used in previous works.

  1. Character Segmentation - obtained from "Tipos Psicológicos", Carl Gustav Jung, 1967. (binary)
  2. Magazine Text Segmentation - obtained from Revista Veja, "Computador - o micro chega às casas", Special Issue, December, 1995. (binary)
  3. Texture Segmentation - obtained from A Short History of the Middle East -- From the Rise of Islam to Modern Times, George E. Kirk, 7th Edition, 1964, Methuen & Co. Ltd., London. (binary)
  4. Boolean noise correction - artificial images. (binary)