XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE KainTenunIkatNet: A new benchmark dataset for Indonesian traditional woven fabric image recognition and image retrieval Silvester Tena 1 Department of Electrical Engineering and Information Technology Universitas Gadjah Mada Yogyakarta, Indonesia 2 Department of Electrical Engineering Universitas Nusa Cendana Kupang, Indonesia siltena@mail.ugm.ac.id Rudy Hartanto Department of Electrical Engineering and Information Technology Universitas Gadjah Mada Yogyakarta, Indonesia rudy@ugm.ac.id Igi Ardiyanto Department of Electrical Engineering and Information Technology Universitas Gadjah Mada Yogyakarta, Indonesia igi@ugm.ac.id Abstract— This study aims to develop the KainTenunIkatNet dataset as one of the benchmarks for image recognition and retrieval problems, especially in Indonesian traditional fabrics. These are due to the present absence of the dataset for traditional fibers, especially the tenun ikat (woven) fabrics of East Nusa Tenggara (ENT). The KainTenunIkatNet dataset contains 120 types of fabric each being captured 40 times, leading to the collection of 4,800 original images at traditional shops and craftsmen. These shooting variations are often internally and externally carried out within the mini-studio box and another fabric background, before being hanged, and worn on the body. In this condition, the utilized Nikon D5600 camera produced 24 megapixels of RGB (red, green, blue) images, with the pre- processing and visual procedural application obtaining a 256x256 display according to the input feature extraction method. Furthermore, the data augmentation processes such as flipping, random rotating, zooming, and shear range, increased the number of images, with each type of tenun ikat fabric being randomly augmented by two images. This led to the production of 9,600 images based on the number of training datasets. Despite this, the testing and validation data were 960 images each. The KainTenunIkatNet dataset is expected to develop image recognition, classification, and retrieval algorithms, while also being useful for the preservation of local culture, as a knowledge base in the fields of education, crafts, and trade. In this report, two pre-trained Convolution Neural Network (CNN) methods with a transfer learning procedure, namely ResNet101 and DenseNet201. The results showed that the retrieval accuracy of two pre-trained models is 100% in the top-1. While the retrieval accuracy of DenseNet201 outperformed the ResNet101 method in the top-5 at 44.75% and 44.38%, respectively. Subsequently, the method had higher retrieval accuracy than ResNet101, although the query time was slower. Keywords— KainTenunIkatNet, feature extraction, transfer learning, image retrieval, CNN I. INTRODUCTION Tenun ikat (woven) fabric is one of the intangible cultural heritages of Indonesia, due to being mostly used at all customary occasions such as birth, wedding, and death ceremonies. This leads to the government's promotion of proposals to UNESCO (United Nations educational, scientific, and cultural organization) and strengthening diplomacy for international intellectual-property protection, through the world and trade rights agreement at the WTO (World Trade Organization). The Tenun ikat fabrics of East Nusa Tenggara also have unique traditional motifs that are in line with local culture. This shows that each region has a different type of motif according to the life of its people. Besides this, they also vary based on animals, flowers, and many geometric shapes [1]. However, the motif users are often difficult to recognize due to the commodity types. The tenun ikat fabric of East Nusa Tenggara also contains the characteristics of color, texture, and shape, although no structured and systematic electronic database has been observed for these commodities. This database is found to provide convenience for craftsmen, education, and trade, with the ease of determining the appropriate motif type and regional origin requiring an image retrieval technique. In the identification of these fabrics, the Content-based image retrieval (CBIR) method is very useful [2], due to overcoming the weakness of the TBIR (text-based image retrieval) model [3]. This explains that CBIR has two components, namely feature extraction and similarity measure methods [4]. The feature extraction method is the most important benchmark in image retrieval, with previous studies on fabric pattern recognition being considerably developed, especially texture characteristics [5], [6]. Furthermore, the CNN method is widely used due to excelling in handling semantic gaps and large datasets, although it utilizes the computational burden [7]. In some previous reports, these pre-trained models were used for object identification, classification, and retrieval [4], [5], [7], [8], [9]. It is also very effective at recognizing objects in multiple dataset fabrics [10], [11], with many reviews developing a new adaptive CNN architecture [12], [13]. These study objectives are observed as follows, (1) Evaluating the KainTenunIkatNet dataset, which contains 9,600 images and 120 classes, as shown in Fig. 1, and (2) Testing the performance of the baseline model for feature extraction (pre-trained CNN), through the transfer learning process. This dataset is useful in the educational field, to introduce the richness of local culture to the younger generation. It is also important for generational craftsmen, to determine the motifs matching their regional origins. In the trade sector, the dataset enables easier electronic determination of motif types and their origin region. It is also useful for developing algorithms in the fields of image recognition, classification, and retrieval, where a feature extraction method is specifically used, i.e., pre-trained CNN. As a benchmark, the KainTenunIkatNet dataset is reportedly analyzed on two pre-trained models, namely DenseNet201 This work was supported by Indonesia Endowment Fund for Education (Lembaga Pengelola Dana Pendidikan Indonesia, ref. 201911212115638). 978-1-6654-7150-3/22/$31.00 ©2022 IEEE 9th ICITACEE 2022 - Semarang, Indonesia, August 25-26, 2022 23