A Bayes decision test for detecting uncovered-background and moving pixels in image sequences.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Detecting changes between image frames is crucial for various applications.
• Existing methods often rely on motion estimation, which can be computationally intensive and error-prone.
• Distinguishing between moving objects and newly uncovered backgrounds is a persistent challenge.

Purpose of the Study:

• To develop a novel ternary hypothesis test for pixel classification.
• To differentiate between stationary, moving, and uncovered-background pixels.
• To avoid the need for explicit motion estimation in the detection process.

Main Methods:

• Utilized the Bayes decision criterion for hypothesis testing.
• Formulated the decision rule based on intensity-difference measurements at individual pixels.
• Incorporated neighborhood intensity-difference measurements to enhance robustness.

Main Results:

• Successfully detected stationary, moving, and uncovered-background pixels.
• Demonstrated that motion estimation is not required for differentiating moving and uncovered-background pixels.
• Evaluated the algorithm quantitatively on a synthetic dataset and qualitatively on a real-world sequence.

Conclusions:

• The proposed ternary hypothesis test offers an effective alternative for pixel change detection.
• The method provides accurate classification without complex motion estimation.
• This approach enhances the efficiency and reliability of image sequence analysis.