Similarity Check Result Implementation of Gabor Filter for Carassius Auratus’s Identification

Abstract— Carassius auratus (goldfish) is a freshwater ornamental fish that is widely kept in both ponds and swimming pools. This fish has various types, colors, and shapes. Several species have similar anatomy so that it is challenging to distinguish manually, including the type of Fantail, Ranchu, Oranda. This study aims to create a system that can identify these species. The identification process uses a feature extraction method, namely the Gabor Filter. 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 addition of pixels for the goldfish image and Gabor's kernel produces a convolution process. The results of the convolution process were normalized to produce a feature vector matrix. The goldfish image classification uses the Probabilitas Neural Network method. The dataset in this research used 216 images of goldfish consisting of the Fantail, Oranda, and Manchu species. The combination of values of each parameter can affect the level of accuracy. Optimal parameters are obtained at kernel size (5.5), frequency (3), orientation (5), and downsample (16.16). The higher the parameter value, the more variation of the feature vector is obtained. The more variations of the feature vector, the higher the data redundancy so that the classification process becomes inefficient. The average accuracy of using a Gabor filter for image identification of goldfish reaches 80.86%.