Displacement field estimation and image segmentation using block matching enhanced by a neural network

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Block matching is crucial for displacement estimation, offering simplicity and noise robustness.
• Accurate displacement field estimation is vital for motion analysis in consecutive images.
• Simultaneous image segmentation and displacement estimation remain challenging.

Purpose of the Study:

• To propose a single-layer feedback neural network for enhanced block matching.
• To simultaneously estimate displacement fields and perform image segmentation.
• To improve the accuracy and efficiency of displacement estimation, especially at object boundaries.

Main Methods:

• A novel single-layer feedback neural network model is developed.
• Modified block matching utilizes embedded segmentation information for accurate displacement fields.
• A flood-fill algorithm enhances dense displacement field computation.

Main Results:

• The proposed model accurately estimates displacement fields and performs image segmentation concurrently.
• Segmentation information embedded in the neural network improves displacement vector accuracy at object edges.
• The flood-fill algorithm provides more efficient and correct dense displacement field computation.

Conclusions:

• The neural network model successfully integrates displacement field estimation and image segmentation.
• Embedding connection relations within the neural network is a novel approach for simultaneous processing.
• The presented methods offer improved accuracy and efficiency in motion analysis and image partitioning.