Convolutional virtual electric field for image segmentation using active contours.

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Gradient Vector Flow (GVF) is effective for active contours but computationally intensive.
• Virtual Electric Field (VEF) offers real-time implementation via Fast Fourier Transform (FFT) as an alternative to GVF.
• Existing models like GVF, VEF, and Vector Field Convolution (VFC) have limitations in segmentation tasks.

Purpose of the Study:

• To introduce the CONvolutional Virtual Electric Field (CONVEF) model as an extension of the VEF model.
• To evaluate CONVEF's performance against GVF, VEF, and VFC models.
• To demonstrate CONVEF's advantages in image segmentation, including enhanced convergence and noise handling.

Main Methods:

• CONVEF models the VEF approach as a convolution operation.
• A modified distance metric is incorporated into the convolution kernel.
• The model leverages Fast Fourier Transform (FFT) for efficient, real-time computation.

Main Results:

• CONVEF exhibits desirable properties like enlarged capture range, u-shape and G-shape concavity convergence, and initialization insensitivity.
• The model effectively separates neighboring objects, suppresses noise, and preserves weak edges.
• Experimental results on synthetic and natural images validate CONVEF's superior performance.

Conclusions:

• CONVEF offers a computationally efficient and robust solution for active contour segmentation.
• The model integrates the benefits of existing methods while introducing novel capabilities.
• CONVEF demonstrates significant potential for real-time image analysis applications.