Symmetry-Based Classification of Regular Textures based on filters bank and Random Forest

Abstract

Texture analysis, a vital component of computer vision and image processing, plays a pivotal role in fields such as decoration and art. This study focuses on the classification of regular textures into 17 distinct wallpaper patterns based on their symmetry operations. Utilizing computer vision techniques and a filter bank approach, we compared three methods: Gabor filter bank, CNN-trained filters, and ImageNet pretrained filters, in conjunction with a random forest model. The results revealed that ImageNet pretrained filters performed exceptionally well, achieving 87% accuracy in the 'wallpaper17' dataset and 81% in the 'wallpaper04' dataset.