An Improved Otsu-Based Image Segmentation Approach for Accurate and Efficient Target Identification
Main Article Content
Abstract
Image processing plays a vital role in enhancing understanding and interpretation, both for humans and computer-based systems. Among its various techniques, segmentation, thresholding, and edge detection are fundamental components in computer vision and image analysis. This paper presents an improved Otsu method for effective segmentation of objects from their image backgrounds. The conventional Otsu method divides an image into background and foreground regions based on a global threshold value. However, it performs poorly for images with a unimodal histogram distribution, as the computed threshold corresponds to the average of the mean intensities of the two classes separated by the threshold value. To overcome this limitation, the proposed improved Otsu method introduces a modification based on weighted background variance. In this approach, the weighting parameter ‘ω’ (ranging from 0 to 1) adjusts the between-class variance, allowing the algorithm to achieve an optimal threshold determined primarily by the variance of the target object. For precise boundary detection of the target, the Laplacian of Gaussian (LoG) operator is applied. The effectiveness of the proposed method is validated through experimental results and compared with other thresholding techniques, including the basic Otsu, Gaussian Otsu, Weighted Object Variance (WOV), and Recursive Otsu methods. The proposed method demonstrates superior performance, achieving an accuracy of 95.46\% for target object detection. It also exhibits faster processing, with an elapsed time of 0.0139 seconds, which is 0.2623 seconds faster than the WOV method and 0.768 seconds faster than the Gaussian Otsu method. Additionally, tests under salt-and-pepper noise (variance = 9) confirm the robustness and reliability of the improved approach for accurate image segmentation.