A Convolutional Neural Network Model for Batik Image Retrieval

The proliferation of digital image data necessitates robust solutions for image retrieval, especially within specialized and culturally significant datasets. This study addresses the inherent challenge of image retrieval within the domain of Batik Nitik, an Indonesian fabric characterized by intricate, hand-drawn designs, where traditional retrieval methods grapple with high-dimensional data and pattern subtlety. We introduce a convolutional neural network (CNN) approach specifically adapted to the complexities of Batik patterns. Our tailored CNN model processes images through a series of convolutional and pooling layers, effectively capturing the distinctive features of Batik designs. The performance of the CNN is evaluated on a dataset comprising 960 Batik Nitik images, employing precision and recall as key metrics, and using distance measures such as Euclidean, Cityblock, Bray Curtis, and Canberra Distance to quantify the accuracy of retrieved images. The proposed model achieved a notable precision of 0.99, significantly outperforming the benchmarks set by established pre-trained models and an Autoencoder in similar scenarios. This high level of precision, coupled with the model's stability reflected in the training and validation phases, underscores its effectiveness. In conclusion, the CNN model designed for Batik image retrieval represents a significant advancement, demonstrating that a specialized approach to feature extraction in image retrieval systems can yield superior results in precision and reliability. © 2024 IEEE.