The specific "randomness" of images of aggregates needs a better understanding, mainly for classification purposes. Such images can be modeled by morphological tools (random set theory) with certain limitations. This paper tries to find a different path by using classical and less classical linear tools and comparing them, in a quantitative or qualitative way. Singular value decomposition plays an important role, as a decorrelation tool, and as a generator for the spectrum of singular values. The "log-profile" of this spectrum is central in this study; in the context of random matrix theory, its properties (approximate linearity...) permit to refer to "pure random situations"; thus, images of aggregates can be understood and modeled as perturbation to these situations. Moreover, independent components analysis corroborates in many cases the preceding results.
[1]
Erkki Oja,et al.
Independent component analysis: algorithms and applications
,
2000,
Neural Networks.
[2]
Anirvan M. Sengupta,et al.
Distributions of singular values for some random matrices.
,
1997,
Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics.
[3]
M. L. Eaton,et al.
The asymptotic distribution of singular values with applications to canonical correlations and correspondence analysis
,
1994
.
[4]
Dominique Jeulin,et al.
Random texture models for material structures
,
2000,
Stat. Comput..