This is SPIRAL-TAP: Sparse Poisson Intensity Reconstruction ALgorithms--theory and practice.

Area of Science:

• Signal Processing
• Statistical Modeling
• Image Reconstruction

Background:

• Many real-world observations involve discrete event counts (e.g., photon detection), necessitating Poisson noise models.
• Conventional penalized least-squares methods are inadequate for reconstructing phenomena from Poisson data.
• Accurate reconstruction is challenging in inverse problems with potentially more unknowns than observations.

Purpose of the Study:

• To develop an effective method for estimating phenomena (f*) from Poisson count data (y) in inverse problems.
• To address the limitations of traditional least-squares approaches for count data.
• To incorporate sparsity and nonnegativity constraints into the estimation process.

Main Methods:

• Utilizes a penalized negative Poisson log-likelihood objective function.
• Implements nonnegativity constraints suitable for Poisson intensity data.
• Employs separable quadratic approximations for iterative optimization.
• Incorporates l1 norms, total variation, and multiscale estimation for regularization.

Main Results:

• The proposed method offers accurate reconstruction of phenomena from Poisson data.
• The approach effectively handles inverse problems with sparse solutions.
• Iterative optimization with separable approximations enhances estimation efficiency.

Conclusions:

• The penalized negative Poisson log-likelihood approach provides a robust solution for inverse problems with count data.
• This method improves the accuracy and efficiency of reconstructing spatially or temporally distributed phenomena.
• The incorporation of sparsity and nonnegativity constraints is crucial for effective Poisson data modeling.