A Multiscale Gradient Fusion Method for Color Image Edge Detection Using CBM3D Filtering

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Traditional edge detection methods struggle with noise and detail loss in color images.
• Existing techniques often fail to preserve fine edge details across different scales.
• Color information is crucial for accurate object boundary identification.

Purpose of the Study:

• To develop a novel color edge detection method integrating collaborative filtering and multiscale gradient fusion.
• To improve edge resolution, noise resistance, and overall image quality in color edge detection.
• To provide a robust alternative to conventional color edge detection algorithms.

Main Methods:

• Utilized Block-Matching and 3D (BM3D) filter for sparse representation enhancement and noise reduction.
• Applied Colored Block-Matching and 3D (CBM3D) filter in the XYZ color space.
• Computed vector gradients and anisotropic Gaussian directional derivatives for refined edge strength mapping.
• Employed normalization, non-maximum suppression, double-thresholding, and morphological refinement for edge enhancement and extraction.

Main Results:

• The proposed method demonstrated robust noise resistance and superior edge quality.
• Experimental results showed improved performance over traditional methods like Color Sobel, Color Canny, SE, and Color AGDD.
• Performance was validated using metrics including the PR curve, AUC, PSNR, MSE, and FOM.

Conclusions:

• The novel color edge detection method effectively integrates collaborative filtering and multiscale gradient fusion.
• The technique significantly enhances edge resolution and overall image quality while maintaining noise resistance.
• The proposed approach represents a substantial advancement in color edge detection technology.