<![CDATA[Segmentation of yellow fish egg spheres in digital images often fails due to the difficulty of determining the boundaries between adjacent or overlapping objects. This research proposes a geometry tracking-based segmentation method to solve the problem. This method uses triangulation of three important edge points around the object to determine the initial segment landmarks. Then, it uses their formation to form a complete circle of candidate segments. The set of pixels enveloped by this circle will be examined for shape and colour to be recognised as segments of an object or not. The method was tested on a fish egg image dataset containing more than 5,473 yellow-orange coloured fish egg spheres in 11 digital images. These egg sphere images vary in size, shape, brightness, contrast, density, shadow, noise, light reflection, and blur. Based on the experimental results, the method was able to correctly segment 4,370 egg spheres with 242 false segments and 1,103 undetected spheres. The performance metrics of this method are precision 94.7%, recall 79.8%, IoU 76.5%, and dice coefficient 86.7%.]]>
Copyrights © 2024
