Statistical processing of large image sequences

Area of Science:

• Remote Sensing
• Image Processing
• Stochastic Processes

Background:

• Dynamic estimation of large-scale stochastic image sequences is crucial for scientific applications.
• Conventional methods like Kalman filters are computationally intensive and impractical for large image datasets.
• Remote sensing applications, such as ocean surface temperature monitoring, generate vast image sequences.

Purpose of the Study:

• To develop a computationally efficient and storage-friendly approach for dynamic estimation of large-scale image sequences.
• To emulate the functionality of Kalman filters with reduced resource requirements.
• To address the challenges posed by nonstationary priors in image sequence modeling.

Main Methods:

• The proposed approach emulates the Kalman filter using a novel static estimation step.
• A mixture of stationary models is employed to accurately mimic nonstationary prior effects.
• The method is demonstrated using a 512 x 512 ocean surface temperature image sequence.

Main Results:

• The developed method significantly reduces computational and storage requirements compared to conventional filters.
• The approach provides an efficient, stable, and positive-definite model.
• The model is consistent with the given correlation structure of the image sequence.

Conclusions:

• The proposed method offers a practical solution for dynamic estimation of large-scale image sequences in remote sensing.
• It simplifies modeling complexity and computational load, making advanced analysis feasible.
• Potential applications include improved modeling and single-frame estimation in various scientific domains.