Classification of Islamic Geometric Patterns based on Machine Learning Techniques

Abstract

Because of their richness and aesthetics, Islamic art and decoration are very popular in many parts of the world. Islamic decoration is characterized by the formation of beautiful visual scenes by putting together simple basic elements. One of the key elements of Islamic style decoration is the use of geometric patterns. In this article, we will focus on classifying Islamic geometric patterns into three symmetrical “frieze, wallpaper, and rosette” groups using machine learning. We use two strategies to describe the Islamic geometric patterns. The first is texture analysis and texture feature extraction, we compare two methods of texture analysis the grey level co-occurrence matrix and Gabor filter bank. The second strategy is to use a pre-trained convolutional neural network to extract features. For classification, we proposed a machine learning classification model based on Random Forest and Support Vector Machine (SVM). The results obtained for the classification of Islamic geometric patterns show that the features extracted using a pre-trained network far outperform the hand-created features, reaching an accuracy of 94%. In the case of hand-created features, the best results were obtained using Gabor filters for feature extraction and random forest for classification with an accuracy of 77%.