"Adaptive Thresholding" Method to Select Prominent Objects by Converting the Image to Black and White Format
Abstract
Image segmentation is a central task in computer vision, pattern recognition, and digital image analysis, as it enables the extraction of meaningful objects from complex backgrounds. However, real-world scenarios often involve non-uniform illumination, noise, and low contrast, which make traditional global thresholding methods, such as Otsu’s algorithm, unreliable. Adaptive Thresholding has emerged as an effective solution by assigning a local threshold to each pixel based on neighborhood statistics. This localized approach enhances object detection in images affected by uneven brightness or distortions. The method has proven useful across diverse domains, including document image analysis, medical diagnostics such as ultrasound and MRI, biometric systems, and intelligent video surveillance. This paper explores Adaptive Thresholding as a strategy for converting grayscale images into binary representations suitable for salient object detection. A theoretical overview of global and adaptive thresholding is provided, followed by mathematical formulations of mean-based and Gaussian-based approaches. Comparative results show that adaptive methods significantly outperform global techniques under challenging lighting conditions. Implementation strategies, algorithm design, and MATLAB experiments are presented, demonstrating improved segmentation accuracy and robustness. Furthermore, practical case studies confirm the effectiveness of adaptive methods in various applications, highlighting their relevance for modern image processing and motivating further research into optimized implementations.