LLPC

LLPC refines manual contour labels by aligning annotated points to image pixel gradient peaks to improve edge localization for supervised deep learning in image segmentation and edge detection.

Key Features:

• Gradient-Guided Point Correction: Adjusts positions of annotated points along object contours by aligning them with pixel gradient peaks to enhance edge localization accuracy.
• Point Interpolation: Interpolates points to address gaps and inconsistencies in contour data.
• Local Linear Fitting Smoothing: Applies local linear fitting-based smoothing to reduce contour roughness and noise-induced irregularities.
• Parameter Efficiency: Operates with only three parameters required for the correction procedure.

Scientific Applications:

• Overlapping Cervical Cell Edge Detection: Used to improve annotation precision for overlapping cervical cell contour delineation.
• Cervical Cell Edge Detection Dataset (CCEDD): Employed to construct CCEDD featuring labels corrected by LLPC for higher precision.
• Performance Improvement: Demonstrated an average precision improvement of 30–40% across multiple networks.

Methodology:

LLPC performs three computational steps: gradient-guided point correction aligning labeled points to pixel gradient peaks; point interpolation to fill gaps and address inconsistencies; and local linear fitting-based smoothing to produce smooth, noise-reduced contours.