An iterative approach to region growing using associative memories

Area of Science:

• Computer Science
• Image Processing
• Hardware Architecture

Background:

• Recursive algorithms are common in image processing but challenging to implement in hardware for real-time applications.
• Traditional recursive region growing methods face difficulties with complex shapes and hardware integration.
• Efficient image segmentation at raster scan rates requires non-recursive approaches.

Purpose of the Study:

• To propose and validate an iterative algorithm for region growing using associative memory.
• To demonstrate the feasibility of implementing recursive computations efficiently through iteration and association.
• To address hardware implementation challenges for real-time image processing tasks.

Main Methods:

• Developed an iterative algorithm for region labeling using content-addressable (associative) memory.
• Designed a hardware structure to support iterative region labeling at scan rates.
• Utilized an equivalence table in hardware to manage pixel region assignments and handle complex shapes.
• Simulated the associative memory for region growing on a serial computer.

Main Results:

• The iterative algorithm successfully performs region labeling efficiently, overcoming limitations of recursive methods.
• Hardware implementation of associative memory enables real-time image segmentation.
• The equivalence table effectively handles complex region shapes (e.g., 'U' and 'N' shapes).
• Simulations confirmed the effectiveness of associative memory for region growing in serial computing.

Conclusions:

• Iterative computation coupled with association provides an efficient alternative to recursion for image processing tasks.
• Associative memory offers a viable hardware solution for real-time image segmentation and region growing.
• The proposed algorithm and hardware structure are suitable for high-speed image analysis.