Identification of Transform Coding Chains.

Area of Science:

• Digital Signal Processing
• Information Theory
• Computer Vision

Background:

• Transform coding is a standard lossy compression technique for discrete sources with memory.
• It involves linear mapping of input vectors, followed by quantization and entropy coding of coefficients.

Purpose of the Study:

• To identify the transform matrix and quantization step sizes in transform coding.
• To address scenarios with limited information, specifically a set of transform decoded vectors.

Main Methods:

• Formulating the problem as finding the lattice with the largest determinant containing observed vectors.
• Proposing and analyzing the convergence properties of an algorithm to find the optimal solution.

Main Results:

• Demonstrating the applicability of the methodology to diverse areas including parameter retrieval, image tampering identification, and predictive coder parameter estimation.
• Validating the proposed approach through experiments on synthetic data and real images.

Conclusions:

• The developed methodology provides a unified framework for identifying transform coding parameters across various applications.
• The lattice-based approach offers an optimal solution with proven convergence for signal compression and analysis.