Two-level global optimization for image segmentation

Abstract Domain-independent image segmentation is considered here as a global optimization problem: to seek the simplest description of a given input image in terms of coherent closed regions. The approach consists of two levels of processing: pixel-level and region-level, both based on the Minimum-Description-Length principle. Pixel-level processing leads to forming the atomic regions that are then labelled. In region-level processing neighbouring regions are merged into larger ones using an explicit attributed graph evolution mechanism. Both level processings are stopped automatically without using any heuristic control parameters. Experiments are carried out with a number of images of different scene types. Parallel implementation of region-level processing is the most difficult problem to be solved for the operational application of this approach.

[1]  Theodosios Pavlidis,et al.  Structural pattern recognition , 1977 .

[2]  D. J. Langridge.  Detection of discontinuities in the first derivatives of surfaces , 1984 .

[3]  Pascal Fua,et al.  Objective functions for feature discrimination: theory , 1989 .

[4]  J. Rissanen,et al.  Modeling By Shortest Data Description* , 1978, Autom..

[5]  Peter J. Rousseeuw,et al.  Robust regression and outlier detection , 1987 .

[6]  Uwe Weidner,et al.  Informationserhaltende Filterung digitaler Bilder und ihre Bewertung , 1991, DAGM-Symposium.

[7]  Wolfgang Förstner,et al.  An MDL-principled evolutionary mechanism to automatic architecturing of pattern recognition neural network , 1992, Proceedings., 11th IAPR International Conference on Pattern Recognition. Vol.II. Conference B: Pattern Recognition Methodology and Systems.

[8]  George Vosselman,et al.  Relational Matching , 1992, Lecture Notes in Computer Science.

[9]  Demetri Terzopoulos,et al.  Regularization of Inverse Visual Problems Involving Discontinuities , 1986, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[10]  Ramesh C. Jain,et al.  Segmentation through Variable-Order Surface Fitting , 1988, IEEE Trans. Pattern Anal. Mach. Intell..

[11]  Wayne Niblack,et al.  Unsupervised image segmentation using the minimum description length principle , 1992, Proceedings., 11th IAPR International Conference on Pattern Recognition. Vol.II. Conference B: Pattern Recognition Methodology and Systems.

[12]  David Lee,et al.  One-Dimensional Regularization with Discontinuities , 1988, IEEE Trans. Pattern Anal. Mach. Intell..

[13]  C. E. SHANNON,et al.  A mathematical theory of communication , 1948, MOCO.

[14]  Andrew Blake,et al.  Visual Reconstruction , 1987, Deep Learning for EEG-Based Brain–Computer Interfaces.

[15]  J. Rissanen.  A UNIVERSAL PRIOR FOR INTEGERS AND ESTIMATION BY MINIMUM DESCRIPTION LENGTH , 1983 .

[16]  Steven W. Zucker,et al.  The Local Structure of Image Discontinuities in One Dimension , 1987, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[17]  Claude L. Fennema,et al.  Scene Analysis Using Regions , 1970, Artif. Intell..

[18]  W. Eric L. Grimson,et al.  Discontinuity detection for visual surface reconstruction , 1985, Comput. Vis. Graph. Image Process..

[19]  M. Nagao,et al.  Edge preserving smoothing , 1979 .

[20]  C. S. Wallace,et al.  A General Selection Criterion for Inductive Inference , 1984, ECAI.

[21]  Martin D. Levine,et al.  Low Level Image Segmentation: An Expert System , 1984, IEEE Transactions on Pattern Analysis and Machine Intelligence.