Motion estimation using higher order statistics

Area of Science:

• Digital Image Processing
• Computer Vision
• Signal Processing

Background:

• Accurate motion estimation is crucial for video analysis.
• Traditional methods struggle with noisy image sequences and small regions.
• Displaced Frame Difference (DFD) is a common metric for motion estimation.

Purpose of the Study:

• Introduce a fourth-order cost function for DFD.
• Improve motion estimation accuracy in severely corrupted image sequences.
• Enable motion estimation for small image regions or blocks.

Main Methods:

• Developed a fourth-order cost function based on DFD statistics.
• Utilized higher-order statistics (beyond second-order) to handle Gaussian noise.
• Compared performance against mean kurtosis and mean square error of DFD.

Main Results:

• The proposed fourth-order cost function demonstrates capability for motion estimation in small regions.
• The method shows effectiveness even when image sequences are severely corrupted by additive Gaussian noise.
• Performance comparisons indicate advantages over traditional mean square error and mean kurtosis methods.

Conclusions:

• Fourth-order statistics offer a robust approach for motion estimation in challenging conditions.
• The novel cost function enhances motion estimation accuracy, particularly for small image blocks.
• This work provides a valuable tool for applications requiring precise motion tracking in noisy video data.