Automatic watershed segmentation of randomly textured color images

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Randomly textured color images present challenges for traditional image processing techniques.
• Accurate segmentation is crucial for automated analysis and defect detection in industrial applications.

Purpose of the Study:

• To develop a new, automatic bottom-up segmentation method for randomly textured color images.
• To introduce a robust algorithm applicable to tasks like automatic granite inspection.

Main Methods:

• A bottom-up segmentation algorithm incorporating color and texture properties.
• Introduction of an LUV gradient for color similarity and watershed transform.
• Merging of watershed mosaic patches based on color contrast and image topology.

Main Results:

• The proposed method effectively segments randomly textured color images without requiring predefined thresholds or markers.
• Demonstrated robustness on granite inspection, yielding good results on diverse noisy color images.
• The algorithm's termination criterion, based on image topology, ensures automatic processing.

Conclusions:

• The developed segmentation method is highly robust and suitable for automatic processing of various noisy color images.
• The LUV gradient and adaptive merging strategy provide an effective approach for texture and color-based image analysis.
• This technique shows significant potential for industrial applications requiring automated image inspection.