Classification of images using Gaussian copula model in empirical cumulative distribution function space

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Traditional image classification often relies on direct pixel value representations.
• Understanding the correlation structure between image features is crucial for improved classification.
• Existing methods may not fully capture contextual relationships within images.

Purpose of the Study:

• To introduce an innovative image classification approach using Gaussian copulas and Empirical Cumulative Distribution Function (ECDF).
• To leverage distribution functions as feature descriptors for enhanced correlation analysis.
• To develop a model that captures contextual relationships for more abstract image representations.

Main Methods:

• Utilized Gaussian copulas combined with an Empirical Cumulative Distribution Function (ECDF) approach.
• Employed Distribution Function of the Distribution Value (DFDV) as the margin distribution.
• Applied Inference Function for Marginals (IFM) principles during the training phase.

Main Results:

• The model achieved an average accuracy of 62.22% on the Modified National Institute of Standards and Technology (MNIST) dataset.
• A peak accuracy of 96.92% was recorded, demonstrating significant performance.
• The use of distribution functions effectively simplified feature description and improved correlation understanding.

Conclusions:

• The proposed Gaussian copula and ECDF-based image classification model shows promising results.
• This method offers a more abstract representation by understanding feature correlations.
• The approach is effective for image classification tasks, as validated by MNIST dataset performance.