Identification of Traditional Motifs Using Convolutional Neural Networks

This paper presents a design for identifying and classifying the Romanian traditional motifs found on 4 different categories (clothes, ceramics, carpets and painted eggs) by training a Convolutional Neural Network (CNN) model derived from the Residual Network (ResNet-50) architecture. We also implemented a system which can detect and identify through a webcam if the object in front of it contains a learned motif. Experimental results show that our neural network has an overall accuracy of 99.4% and a reduced webcam processing time.