A new iterative triclass thresholding technique in image segmentation.

Area of Science:

• Computer Vision
• Image Processing
• Pattern Recognition

Background:

• Standard Otsu's method segments images based on a global threshold.
• Otsu's method can struggle with complex images containing fine structures or weak objects.

Purpose of the Study:

• To develop an iterative image segmentation technique that enhances Otsu's method.
• To improve the accuracy of image segmentation for challenging visual data.

Main Methods:

• An iterative approach that refines subregions of an image for segmentation.
• The method initially applies Otsu's threshold, then iteratively segments the 'to-be-determined' (TBD) regions.
• Segmentation ceases when the difference between successive Otsu's thresholds falls below a predefined limit.

Main Results:

• The iterative method successfully segmented challenging synthetic and real-world images.
• Demonstrated superior performance compared to standard Otsu's method in identifying weak objects and fine details.
• Achieved enhanced segmentation accuracy with only a minimal increase in computational cost.

Conclusions:

• The proposed iterative Otsu's method offers improved image segmentation capabilities.
• This technique is effective for complex image analysis tasks requiring high detail.
• The method provides a computationally efficient enhancement over traditional Otsu's segmentation.