Improving the segmentation of digital images by using a modified Otsu's between-class variance

Area of Science:

• Computer Vision
• Image Processing

Background:

• Image segmentation is crucial for image analysis, involving pixel division based on intensity.
• Multi-level thresholding (MT) offers advantages over bi-level methods but faces computational complexity.
• Otsu's method and Kapur's entropy are established techniques for image segmentation thresholding.

Purpose of the Study:

• To propose a novel modified Otsu function for image segmentation.
• To hybridize Otsu's between-class variance with Kapur's entropy for enhanced thresholding.
• To evaluate the performance of the proposed method using metaheuristic algorithms.

Main Methods:

• A modified Otsu function is developed by integrating Otsu's between-class variance and Kapur's entropy maximization.
• The Arithmetic Optimization Algorithm (AOA) and Hybrid Dragonfly-Firefly Algorithm (HDAFA) are employed for optimization.
• The proposed method is tested on images from the Berkeley Segmentation Data Set 500 (BSDS500).

Main Results:

• The modified Otsu technique effectively combines maximum variance and entropy for optimal threshold selection.
• Experimental results demonstrate superior performance compared to standard Otsu and Kapur methods.
• The approach shows high efficiency in metrics like Peak Signal-to-Noise Ratio (PSNR) and segmentation quality.

Conclusions:

• The proposed modified Otsu method offers an efficient and accurate solution for image segmentation.
• Hybridizing Otsu's variance with Kapur's entropy improves thresholding by considering both variance and uncertainty.
• The method shows significant potential for various image processing applications.