The paper presents a novel method for automated image processing that combines pyramidal generalized contour preprocessing with parallel–hierarchical segmentation, integrating adaptive multilevel thresholding to enhance segmentation accuracy and robustness. The proposed approach is designed to overcome the limitations of traditional methods—whose performance declines under variations in brightness, surface texture, and noise—by enhancing image contrast and structural defect detection, thereby reducing diagnostic errors and misclassification risks. To achieve these objectives, the implementation utilizes multilevel adaptive thresholding, enabling step-by-step segmentation refinement and the extraction of informative regions using three-level coding (positive, negative, and neutral elements). In conjunction with parallel–hierarchical (PH) transformations and high-frequency filtering, the method enhances image contrast, enables more accurate detection of structural defects, and reduces the number of false positives. Experimental results demonstrate a 10–15% improvement in segmentation accuracy compared to classical methods such as region-growing techniques. Furthermore, correlation analysis between automatic and manual segmentation results demonstrated a high degree of consistency, with a correlation coefficient of 0.95–0.99, indicating the reliability and reproducibility of the developed approach. The proposed method is distinguished by its high processing speed, computational simplicity, and versatility of application, ranging from medical thermography for pathological diagnostics to real-time monitoring of railway infrastructure. The practical significance of these results lies in advancing automation, reducing decision-making errors, and ensuring greater reliability of technical and medical control systems.
Lukoševičius et al. (Fri,) studied this question.