Goldfish (Carassius auratus) Segmentation Using Expectation Maximization (EM) Method

Abstract

Carassius Auratus, also known as goldfish, is the most commonly kept ornamental fish in an aquarium that has a variety of species, shapes, and colors. Goldfish anatomy is almost similar between species so it is difficult to distinguish. Goldfish identification has several processes to determine the level of success. One process that is very important for identification is segmentation. Segmentation is the stage to separate objects and backgrounds. Good segmentation results will provide maximum feature extraction results and it should impact on optimal identification results. This research aims to create a system that can be used to segment goldfish fish objects in the background using the Expectation-Maximization method. Expectation-Maximization is an algorithm for estimating a parameter in a function by using Maximum Likelihood Estimation (MLE). Goldfish fish species total of 72 images each totaling 216. Goldfish segmentation evaluation results using the Expectation-Maximization method can work well with an accuracy rate of 89.14%. Analysis of the results of the imperfect fish image segmentation is influenced by white light and the background color is almost similar to that of the goldfish. The results of the best goldfish image segmentation are influenced by a good goldfish image when the image capture process at the time of the screenshot has a stable focus and is not blurry so that the edges of the goldfish object appear clearly and sharply.