On the modeling of small sample distributions with generalized Gaussian density in a maximum likelihood framework

Area of Science:

• Signal processing
• Statistical modeling
• Image analysis

Background:

• Generalized Gaussian density (GGD) is widely used for modeling sample distributions.
• Existing methods often assume large sample sizes for parameter justification.
• The behavior of GGD parameters on small samples remains underexplored.

Purpose of the Study:

• To investigate the differences in GGD parameter computation between small and large samples.
• To establish a condition for the existence of GGD parameters in a maximum likelihood framework.
• To develop and compare a new algorithm for GGD parameter estimation.

Main Methods:

• Utilizing a maximum likelihood estimation framework.
• Deriving a necessary and sufficient condition for GGD parameter existence.
• Developing a novel algorithm for parameter computation.
• Comparing the proposed algorithm with existing methods on various image sizes.

Main Results:

• Demonstrated that GGD parameter computation is sample-size dependent.
• Established a clear condition for the existence of GGD parameters.
• The derived algorithm shows competitive performance compared to existing methods, especially on smaller datasets.
• The study highlights the limitations of asymptotic assumptions for small samples.

Conclusions:

• The accurate estimation of generalized Gaussian density parameters is crucial and sample-size dependent.
• The proposed condition and algorithm offer a more robust approach to GGD parameter estimation, particularly for small samples.
• This work provides a valuable contribution to statistical modeling and image analysis techniques.