Content-based Traditional Clothes Pattern Retrieval Using Convolutional Neural Network

Indonesia is a country with an inclusive culture spread all over the country. Usually, cultural objects can have different meanings based on the details. Traditional cloth is an example of Indonesia's culture. Each province in Indonesia has various colors, patterns, textures, characteristics of traditional fabrics. However, each cloth's patterns and textures have different meanings or meanings. The various features and characteristics of traditional clothing can be extracted using the Convolutional Neural Network. This experiment uses 74 types of traditional patterns collected from the internet. Pre-processing creates 4144 new images with seven different rotation variations, four different overlap settings, and also two crop sides for the image database. CNN type for VGG19 is used as feature extraction of fabric pattern and Euclidean Distance for feature vector distance calculation. Based on the experimental results, the average output using this method is 86%. Besides that, about 69% of the results of the query process for pattern and texture images have an accuracy greater than 80%. Even some patterns can produce accuracy up to 100%. This study makes an excellent image retrieval accuracy but also needs a lot of improvement.