Siti Fatimah, Sony Hartono Wijaya
Fishes are the most numerous and diverse of the major vertebrate groups causing difficulties in fish image recognition. Therefore, feature extraction in content-based image retrieval is needed to obtain pattern or fish features. This research uses freshwater fish images categorized based on the family. This research started with collected fish images from Axelrod’s Mini-Atlas of Freshwater Aquarium Fishes Mini-Edition. The first step in this research is segmented fish images with expectation maximization algorithm for grouping color contained into images. Next step is image extraction method by using three image visual features (color, shape, texture). Color feature extraction with Fuzzy Color Histogram (FCH) method is used through computing the membership function using Cauchy function. Shape feature is extracted using Hough Transform Circle and Co-occurrence Matrix method is computed for texture feature extraction. Similarity value between image query and images in database is then computed based on its features (color, shape, and texture) using cosine similarity. Those features were combined with Bayesian Network Model. Average precision value on all families fish images database for color, shape, texture and bayesian network are 0.2586, 0.2589, 0.2479, 0.2639 respectively. However, they don’t significantly different.
Keywords: fishes, feature extraction, content-based image retrieval (CBIR), fuzzy color histogram, hough transform circle, bayesian network model