Gradient Vector Flow (GVF)

Pure MATLAB version of Quan Wang's Fast Gradient Vector Flow (GVF).

This implementation was originally based on Quan Wang's Fast Gradient Vector Flow (GVF): https://www.mathworks.com/matlabcentral/fileexchange/45896-fast-gradient-vector-flow-gvf

On tested hardware, this pure MATLAB implementation achieved comparable or better performance than the referenced C++ implementation.

Implementation details:

Compared to the original implementation, this MATLAB version introduces several optimizations focused on computational efficiency and memory usage:

• The Laplacian term is computed using a direct 4-neighbour finite-difference stencil implemented via conv2(), avoiding the overhead of del2() and providing approximately a 3x speed improvement over del2().
• Computations are performed in single precision, reducing memory consumption and improving runtime by approximately 2x compared to double precision.
• Memory usage is reduced by minimizing temporary array allocations during the iterative update process. The Laplacian buffer is reused for both vector field components.
• Boundary conditions are handled efficiently by updating the existing one-pixel symmetric padding instead of repeatedly creating padded copies of the image.
• The implementation supports CUDA-enabled GPUs through MATLAB's gpuArray framework. Performance improvements depend on image size and GPU hardware.