International Journal on Advanced Science, Engineering and Information Technology, Vol. 11 (2021) No. 2, pages: 566-571, DOI:10.18517/ijaseit.11.2.8128

Implementation of Gabor Filter for Carassius Auratus’s Identification

- Aristoteles, Yunda Heningtyas, Admi Syarif, AA Gieniung Pratidina


Carassius Auratus, otherwise known as goldfish, is one of the most popular ornamental fish. Goldfish have many variations, such as differences in body shape, colors, size, and fins. Identifying goldfish manually is difficult to do. This is due to several species that have similar anatomy, so automatic fish identification is needed. This research aims to identify three species of goldfish, such as Fantail, Oranda, and Ranchu. Gabor filter was applied to extract the features of goldfish. Gabor filter consists of several steps, including parameter initialization, Gabor kernels, Gabor convolution, feature point. The parameters used were frequency, orientation, and kernel’s size. Gabor kernel was formed based on initialized parameters. The convolution process was produced by adding up the multiplication of 256x256 pixel goldfish’s images and Gabor kernels. The results of the convolution process were normalized to produce a feature vector matrix. A probability neural network was used to classify the goldfish. Probability Neural Network is a supervised network that finds its natural use in decision making and classification problems. This research used 216 of goldfish’s images. Seventy-two images were used for each species. The optimal parameters in this study were kernel size (5,5), frequency (3), orientation (5), and downsample (16,16), with accuracy up to 100%. Parameters of the frequency, orientation, kernel size, and downsample affect the level of accuracy. The greater the parameter value used, the more variations in feature vectors are obtained. Still, if too many variations of the feature vector cause redundancy data, it causes the classification process to be inefficient.


Extraction feature; gabor filter; goldfish identification; pattern recognition; probability neural network.

Viewed: 1136 times (since abstract online)

cite this paper     download