To FRAME or not to FRAME in probabilistic texture modelling?

The maximum entropy principle is a cornerstone of FRAME (filters, random fields, and maximum entropy) model considered at times as a first-ever step towards a universal theory of texture modelling or even as "the inevitable texture model". This paper disputes such opinions. That a wealth of exponential families of probability distributions is deduced from the ME principle is well known for decades. The ME properly by itself in no way leads to an adequate probabilistic description, and to model a particular texture, specific limitations have to be imposed on signal statistics. Frequency distributions of outputs from a bank of linear filters (the second FRAME'S cornerstone) are hardly the only choice outperforming all other alternatives. The paper points also to other hidden drawbacks of FRAME.

[1]  Song-Chun Zhu,et al.  Modeling Visual Patterns by Integrating Descriptive and Generative Methods , 2004, International Journal of Computer Vision.

[2]  Song-Chun Zhu,et al.  Equivalence of Julesz and Gibbs texture ensembles , 1999, Proceedings of the Seventh IEEE International Conference on Computer Vision.

[3]  Stan Z. Li,et al.  Markov Random Field Modeling in Computer Vision , 1995, Computer Science Workbench.

[4]  L. Gool,et al.  Parallel composite texture synthesis , 2002 .

[5]  K. Pahlavan,et al.  Texture Modeling by Multiple Pairwise Pixel Interactions , 1996 .

[6]  S. Kullback,et al.  Information Theory and Statistics , 1959 .

[7]  Rupert Paget,et al.  Texture synthesis via a noncausal nonparametric multiscale Markov random field , 1998, IEEE Trans. Image Process..

[8]  Andrew Zisserman,et al.  Texture classification: are filter banks necessary? , 2003, 2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2003. Proceedings..

[9]  Robert G. Aykroyd,et al.  Bayesian Estimation for Homogeneous and Inhomogeneous Gaussian Random Fields , 1998, IEEE Trans. Pattern Anal. Mach. Intell..

[10]  Song-Chun Zhu,et al.  Statistical Modeling and Conceptualization of Visual Patterns , 2003, IEEE Trans. Pattern Anal. Mach. Intell..

[11]  Song-Chun Zhu,et al.  Exploring Texture Ensembles by Efficient Markov Chain Monte Carlo-Toward a 'Trichromacy' Theory of Texture , 2000, IEEE Trans. Pattern Anal. Mach. Intell..

[12]  Gerhard Winkler,et al.  Image analysis, random fields and dynamic Monte Carlo methods: a mathematical introduction , 1995, Applications of mathematics.

[13]  Song-Chun Zhu.  Filters, Random Fields and Maximum Entropy (FRAME): Towards a Unified Theory for Texture Modeling , 1998 .

[14]  B. O. Koopman.  On distributions admitting a sufficient statistic , 1936 .

[15]  John P. Moussouris.  Gibbs and Markov random systems with constraints , 1974 .

[16]  Eero P. Simoncelli,et al.  A Parametric Texture Model Based on Joint Statistics of Complex Wavelet Coefficients , 2000, International Journal of Computer Vision.

[17]  Georgy L. Gimel'farb,et al.  Image Textures and Gibbs Random Fields , 1999, Computational Imaging and Vision.

[18]  Glenn Healey,et al.  Modeling and classifying color textures using random fields in a random environment , 1999, Pattern Recognit..

[19]  Georgy L. Gimel'farb,et al.  Low‐level bayesian segmentation of piecewise‐homogeneous noisy and textured images , 1991, Int. J. Imaging Syst. Technol..

[20]  Song-Chun Zhu,et al.  Minimax Entropy Principle and Its Application to Texture Modeling , 1997, Neural Computation.

[21]  Luc Van Gool,et al.  A Compact Model for Viewpoint Dependent Texture Synthesis , 2000, SMILE.

[22]  Luc Van Gool,et al.  Multiview texture models , 2001, Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001.

[23]  P. Spreij.  Probability and Measure , 1996 .