Putra Wisnu Agung Sucipto, Annisa Firasanti, Eki Ahmad Zaki Hamidi
This study proposes an alternative way to segment fish egg images, using label propagation techniques with a vertical and horizontal proximity approach when dealing with segments that have isolated pixels. This work was tested on 5,473 yellow-orange fish eggs measuring 2-4 mm, varying in size, shape, brightness, contrast, density, shadow, noise, light reflection, and blur. Testing was carried out by setting the horizontal (wh), vertical (wv), and diagonal (wd) direction weight parameters in the range of 0.0001 to 2000, and a fixed cluster density (σeff = 1). The test results showed a reduction in isolated pixels, with an accuracy of 31.34, precision of 0.54, the F1-score of this method (0.63), recall (0.76), and MAE of 362.82. This work has shown the accuracy to be able to approach the ground truth, proven to be effective and adaptive for segmenting complex images with small and scattered objects. © 2025 IEEE.
State University of Malang, Department of Electrical and Informatic Engineering, Malang, Indonesia; University of Islamic 45, Electrical Engineering Department, Bekasi, Indonesia; Uin Sunan Gunung Djati Bandung, Electrical Engineering Department, Bandung, Indonesia
