ความฉลาดเชิงกลุ่มในอัลกอริทึมการตรวจจับขอบที่ไม่ชัดเจน

Abstract

Medical images have ill-defined edges caused by noise, therefore, it is difficult to identify edge boundaries. In the literature, the ill=defined edge detection [1] is very effective in finding such edges. However, the parameters used in the algorithm have to be selected manually. In this thesis, we utilized three swarm intelligence algorithms, i.e., Particle Swarm Optimization (PSO), Aquila Optimizer (AO), and Modified Marine Predators Algorithm (MMPA) to find suitable parameters. We evaluate this proposed algorithms on five data sets, i.e., synthetic images with Gaussian noise, synthetic images with uniform noise, wrist X-ray images, knee CT scan images, and prostate ultrasound images. The Intersection over Union (IOU) metric was used to evaluate the performance of the proposed method. The results showed that the proposed method was able to effectively find optimal parameters for edge following, with average IOU scores of 0.9197±0.01511, 0.93153±0.01218, 0.92575±0.05121, 0.88134±0.10417, and 0.50240±0.22601 on the five datasets, respectively. The main source of errors in edge detection was found to be the discontinuity of the average edge vector field and edge map, as well as noise in the starting point, which can lead to incorrect identification of edge pixels.